A coordinated post-quantum mesh and route-aware zero-trust framework for smart-grid IoT communication utilizing simulation and attack evaluation

1 Introduction

The continuing digital transformation of the energy sector has accelerated the deployment of smart grids, advanced metering infrastructure (AMI), distributed energy resources (DERs), substation automation, intelligent sensors, and Internet of Things (IoT)-enabled monitoring and control systems. These technologies support demand-response management, distributed generation coordination, fault monitoring, load forecasting, and real-time operational decision-making. However, the increasing connectivity among smart meters, DER controllers, field routers, substations, and utility control centres also expands the cyber-attack surface. Smart-grid communication infrastructures are exposed to man-in-the-middle (MITM), distributed denial-of-service (DDoS), replay, spoofing, selective forwarding, routing manipulation, false-data injection, and unauthorized-access attacks (Grice et al., 2025; Gungor et al., 2012; Wang and Lu, 2013; Tan et al., 2024; Mejia-Ruiz et al., 2025). A successful attack may disrupt data availability, compromise control commands, reduce operational visibility, or delay the delivery of time-sensitive grid information. For AMI, such disruption can delay billing and demand visibility; for DER supervision, it can produce stale active/reactive-power set-points; and for substation monitoring, it can delay equipment-state and alarm information. These consequences motivate application-specific communication-security targets without implying that communication results alone prove electrical-system resilience. Securing these infrastructures is challenging because smart-grid communication networks are heterogeneous, geographically distributed, latency-sensitive, and frequently composed of resource-constrained devices. Many field devices have limited processing capacity, memory, bandwidth, and energy reserves, while remaining operational for 10–20 years or longer (Gungor et al., 2012; Wang and Lu, 2013). Centralized security architectures may provide effective authentication and access control under normal conditions, but their dependence on a limited number of gateways or authentication servers can create communication bottlenecks and single points of failure. Security mechanisms for future energy systems must therefore maintain confidentiality, integrity, authentication, availability, and routing resilience without imposing unacceptable computational or communication overhead. A related long-term concern is the potential impact of cryptographically relevant quantum computers on classical public-key cryptography. RSA and elliptic-curve cryptography (ECC) depend on integer-factorization and discrete-logarithm problems that are computationally difficult for classical computers but theoretically vulnerable to Shor’s algorithm. This creates a harvest-now, decrypt-later risk in which an adversary stores encrypted smart-grid traffic and attempts to decrypt it when sufficiently capable quantum computers become available. This threat is particularly relevant to long-lifetime grid assets, archived metering records, infrastructure credentials, and remotely deployed controllers that cannot be replaced frequently (Shekhawat and Gupta, 2024; Asif, 2021; Chawla and Mehra, 2023; Saad Ahmed et al., 2026).

The National Institute of Standards and Technology standardized ML-KEM for post-quantum key establishment and ML-DSA for post-quantum digital signatures. These mechanisms provide a practical foundation for quantum-safe migration; however, their larger public keys, ciphertexts, and signatures can increase memory demand, IEEE 802.15.4 fragmentation, verification delay, and radio energy (Jabłoński and Dylewski, 2025; Shekhawat and Gupta, 2024; Asif, 2021; Bera and Sikdar, 2024; Chawla and Mehra, 2023). Section 4.2 therefore compares ML-KEM-512/768/1024, ML-DSA-44/65/87, and a hybrid P-256–ML-KEM profile on the same Raspberry Pi platform. Post-quantum migration is consequently assessed through latency, memory, energy, delivery, throughput, control overhead, routing stability, and attack recovery rather than cryptographic strength alone. Existing work can be grouped into PQC-focused smart-grid authentication, secure or trust-based RPL, endpoint-centred zero trust, and attack-recovery or resilient-routing studies. These streams provide useful mechanisms but generally do not bind standardized PQC identity to route admission, continuous path trust, isolation, authenticated failover, and session rekeying in one controlled workflow. Moreover, an endpoint comparison between a classical centralized system and a complete PQC mesh cannot identify whether an observed change results from cryptography, topology, routing behaviour, or zero-trust policy. A component-wise and interaction-based evaluation is therefore required. To address this problem, this study presents a coordinated post-quantum, RPL-based mesh, and route-aware zero-trust security architecture for smart-grid IoT communication. The framework uses ML-KEM-1024 for session-key establishment, ML-DSA-87 for device and routing-message authentication, and authenticated RPL control messages for decentralized route formation. The principal technical contribution is not a new cryptographic primitive or routing protocol; rather, it is a coordinated security workflow that binds post-quantum device authentication to route admission and continued network participation.

The workflow incorporates PQC-authenticated node admission, route-level trust scoring, continuous route revalidation, signed routing-control messages, trust-triggered node isolation, multipath failover, session-key refresh following route or trust-state changes, and controlled node recovery and reintegration. A node is admitted only after successful identity and signature verification. Its subsequent routing behaviour, packet-forwarding reliability, message integrity, and authentication status contribute to a continuously updated trust state. When the trust state falls below the defined acceptance threshold, the node or affected route is isolated, an authenticated alternative path is activated, and the secure session is refreshed. To the authors’ knowledge, limited work has evaluated the coordinated operation of standardized PQC, RPL-based mesh resilience, and route-aware zero-trust verification through a controlled multi-configuration smart-grid study. The proposed framework is evaluated using five configurations: classical centralized communication, classical mesh communication, post-quantum centralized communication, post-quantum mesh communication, and the complete post-quantum mesh architecture with zero-trust validation. In addition, a published secure-RPL or trust-routing framework, ETX–Trust Secure-RPL baseline, is reproduced under the same environment to provide a direct comparison beyond the proposed configurations. This design isolates the effects of cryptography, topology, and continuous trust enforcement under normal traffic, high-load operation, MITM attacks, DDoS attacks, route manipulation, and network-failure conditions. The evaluation is grounded in a distribution-grid communication scenario comprising smart meters, DER controllers, substation monitoring devices, field routers, gateways, and a utility control centre. Communication metrics are interpreted in relation to defined grid functions, including meter telemetry, DER supervisory control, substation condition monitoring, and demand-response command delivery. Quantum-security assurance is assessed using standardized parameter sets and NIST security categories, while migration costs are examined through a transparent scenario-based economic model.

1.1 Research questions

This study is guided by four questions directly mapped to the evaluation evidence:

• RQ1: How can ML-KEM and ML-DSA be coordinated with RPL route admission and route-aware zero trust, and which session, route-admission, freshness, failover, and rekeying properties are supported by symbolic verification and attack tests?
• RQ2: What are the individual effects of PQC and mesh topology on RTT, energy, CPU use, PDR, throughput, and control overhead across C1–C4, and how does C5 compare with the reproduced published secure-RPL or trust-routing baseline under identical conditions?
• RQ3: What additional detection, isolation, and recovery effects arise from route-aware zero trust in C4, C4-T, C3-ZT, and C5 under MITM, replay, route manipulation, selective forwarding, DDoS, and router failure?
• RQ4: What security, parameter-set, interoperability, standards-integration, and scenario-based economic trade-offs follow from the selected migration profile, and what validation boundaries remain?

1.2 Major contributions

The principal contributions of this study are summarized as follows:

• A coordinated protocol-level workflow integrating ML-KEM-1024, ML-DSA-87, signed RPL route admission, continuous route trust, isolation, multipath failover, session-key refresh, and controlled reintegration; no new cryptographic primitive or routing algorithm is claimed.
• A controlled C1–C5 evaluation, supplemented by C4-T and C3-ZT targeted ablations and a same-environment reproduction of ETX–Trust Secure-RPL baseline, to distinguish cryptographic, topological, trust-scoring, isolation, recovery, and external-baseline effects.
• A distribution field-area communication case study linking RTT, delivery, recovery, and availability to AMI, DER supervisory-control, demand-response, and substation-monitoring communication targets.
• A reproducible methodology combining independently seeded NS-3 runs, Raspberry Pi measurements, attack-specific reporting, sensitivity analysis, formal protocol verification, confidence intervals, effect sizes, interaction tests, and a public repository containing the code, configurations, seeds, traces, formal model, and analysis files at [insert the already-posted repository name and persistent URL/DOI from the repository record].
• A standards-based quantum-security interpretation, a dedicated integration mapping for IEEE 802.15.4/6LoWPAN, RPL/MRHOF, CoAP/UDP, IEC 62351, IEC 61850, IEEE 2030.5, DNP3 Secure Authentication, and applicable NERC-CIP practices, and a scenario-based economic assessment with conservative, baseline, optimistic, and Monte Carlo uncertainty outputs.

The remainder of this article is organized as follows. Section 2 critically reviews post-quantum smart-grid security, RPL mesh resilience, zero-trust architectures, DER communication requirements, and cyber-physical co-simulation. Section 3 defines the smart-grid system model, communication requirements, and threat model. Section 4 presents the proposed PQC–mesh–zero-trust protocol and coordination workflow. Section 5 describes the evaluation methodology, five system configurations, attack procedures, performance metrics, and statistical methods. Section 6 presents the quantitative results, while Section 7 interprets their smart-grid operational and security implications. Section 8 presents the scenario-based migration and economic analysis. Section 9 discusses limitations and future validation, and Section 10 concludes the study.

2 Literature review

Research relevant to quantum-safe smart-grid communication spans post-quantum cryptography, low-power mesh routing, zero-trust access control, DER cybersecurity, and cyber-physical co-simulation. Although these areas have individually progressed, their assumptions and validation methods differ considerably. This section therefore examines not only the mechanisms proposed in previous studies but also their energy application, network architecture, attack coverage, grid-operation representation, reproducibility, and principal limitations.

2.1 Post-quantum cryptography for smart-grid and IoT communication

Post-quantum security research for smart grids has investigated lattice-based encryption, quantum-resistant authentication, secure data aggregation, quantum key distribution, and hybrid cryptographic migration (Grice et al., 2025; Jabłoński and Dylewski, 2025; Li et al., 2019; Shekhawat and Gupta, 2024; Asif, 2021; Qian et al., 2021; Prateek et al., 2024; Xiong et al., 2025; Bera and Sikdar, 2024; Chawla and Mehra, 2023; Kong, 2020). These studies establish that long-lifetime smart meters, gateways, and grid controllers require protection against harvest-now, decrypt-later attacks and future compromises of RSA- and ECC-based credentials.

Lattice-based mechanisms are particularly relevant because ML-KEM and ML-DSA provide standardized approaches to key establishment and digital signatures. Nevertheless, their suitability for energy IoT systems depends on more than nominal security strength. Larger public keys, ciphertexts, and signatures can increase packet fragmentation, memory use, verification delay, and radio-energy consumption. Consequently, PQC performance must be evaluated with explicitly defined parameter sets, software implementations, device capabilities, packet formats, and network traffic conditions. Existing cryptographic studies provide valuable implementation evidence, but many evaluate cryptographic operations independently of routing changes, compromised-node isolation, and smart-grid application requirements (Jabłoński and Dylewski, 2025; Shekhawat and Gupta, 2024; Asif, 2021; Bera and Sikdar, 2024; Chawla and Mehra, 2023).

Hybrid classical/PQC operation is also relevant during migration because utilities may require dual credentials, rollback paths, crypto-agility, and staged certificate replacement (Xiong et al., 2025; Chawla and Mehra, 2023). The present full-PQC configurations isolate post-quantum overhead, while the hybrid P-256–ML-KEM microbenchmark represents the transitional case without changing the principal C1–C5 design.

2.2 Mesh and RPL security in low-power energy networks

RPL is widely used to organize IPv6-based communication in low-power and lossy networks. Its distributed structure can provide alternative paths, reduce reliance on a single gateway, and improve communication continuity during link or node failure. These properties are attractive for AMI, DER monitoring, and geographically distributed field devices (Faheem et al., 2024; Gungor et al., 2012; Wang and Lu, 2013; Tan et al., 2024).

However, RPL is vulnerable to rank manipulation, sinkhole formation, selective forwarding, replayed control messages, and malicious parent selection (Hassan et al., 2026, 2024). Mesh connectivity alone therefore does not provide a trusted routing environment. Resilient operation requires authenticated routing-control messages, verifiable node identities, route-quality monitoring, and procedures for isolating suspicious nodes. Previous mesh studies commonly emphasize packet delivery and fault tolerance, while cryptographic identity, post-quantum migration, and continuous route-level trust are less frequently evaluated together (Hassan et al., 2026, 2024).

Secure-RPL extensions, trust-based parent selection, and intrusion-detection-assisted routing generally authenticate control traffic or score nodes, but they do not necessarily bind PQC-authenticated identity, minimum path trust, automatic quarantine, failover, and rekeying (Hassan et al., 2026, 2024). To provide a direct external comparison rather than only an internal configuration study, the published framework identified as ETX–Trust Secure-RPL baseline is reproduced under the same topology, traffic, seeds, attack schedule, and metric definitions. Any deviations required to reproduce that framework in NS-3 are documented explicitly in Section 5.

2.3 Zero-trust architectures for cyber-physical and operational-technology systems

Zero-trust security replaces implicit network trust with continuous verification of identities, devices, sessions, and access requests. In smart-grid and operational-technology environments, zero-trust principles have been applied to terminal protection, fog-assisted authentication, smart-meter access, and network micro-segmentation (Tao et al., 2023; Prateek et al., 2024). These approaches can restrict lateral movement and reduce the impact of compromised credentials.

Nevertheless, conventional zero-trust implementations often concentrate on users, applications, or gateways rather than the trustworthiness of distributed routing paths. Smart-grid mesh networks require route-aware enforcement because an authenticated device may subsequently behave maliciously by dropping packets, advertising deceptive routing information, or forwarding traffic through an untrusted path. The present study therefore extends continuous verification from device admission to route participation by combining cryptographic identity validation with routing reliability, message-integrity history, behavioural compliance, and trust-triggered failover (Hassan, 2026).

Endpoint trust determines whether a device or session is admitted, whereas route trust evaluates the continuing behaviour and minimum trustworthiness of all forwarding nodes on an active path. This distinction supports route-triggered isolation and recovery rather than access control alone.

2.4 DER, AMI, SCADA, and real-time grid communication requirements

Smart-grid cybersecurity must be evaluated in relation to the operational functions carried by the communication network (Gungor et al., 2012; Wang and Lu, 2013; Tan et al., 2024; Mejia-Ruiz et al., 2025). AMI traffic generally supports periodic meter readings, event reporting, outage detection, and demand-response signalling. DER communication supports inverter monitoring, active and reactive power set-points, battery coordination, and voltage-support functions. SCADA and substation communication provide supervisory control, state monitoring, alarm delivery, and operational visibility.

The survey of cybersecurity challenges in power networks with distributed energy resources (Mejia-Ruiz et al., 2025) demonstrates that cyber incidents can affect more than communication confidentiality. Compromised or delayed DER information may degrade voltage regulation, distort system-state estimates, interrupt coordinated dispatch, or prevent timely corrective actions. This implies that communication metrics should be interpreted according to the grid function they support. A packet delay acceptable for periodic meter collection may be unsuitable for a time-sensitive DER control command, while temporary packet loss may have different consequences for billing data and substation alarms.

Accordingly, the present study maps network latency, packet-delivery ratio, availability, and recovery time to specific AMI, DER, demand-response, and substation-monitoring traffic profiles. This avoids treating all smart-grid packets as having identical operational importance.

2.5 Power–communication co-simulation and cyber-physical validation

Network-only simulation is useful for isolating cryptographic overhead, routing behaviour, packet loss, and attack recovery. However, cyber-physical validation requires an additional power-system model to determine how communication disturbances influence electrical variables and control actions.

Real-time transmission–distribution co-simulation with DER-based voltage and frequency support (Mejia-Ruiz et al., 2023) shows that DER coordination depends on timely information exchange between transmission operators, distribution controllers, and distributed resources. Communication delay, message loss, or control-channel disruption can therefore influence the availability and timing of voltage and frequency support.

These findings provide two important implications for security evaluation. First, cyber metrics should be connected to the timing and reliability requirements of the associated grid function. Second, power–communication co-simulation provides a stronger subsequent validation stage for determining whether a communication-level disturbance produces a measurable electrical consequence. The present evaluation therefore focuses on the communication-security layer while defining operationally meaningful traffic profiles that can later be connected to real-time transmission and distribution models.

2.6 Research gap and positioning of the present study

Existing research provides strong individual treatments of PQC, decentralized routing, and zero-trust security, but lacks a clearly specified coordination workflow and controlled component-wise assessment in a concrete smart-grid communication scenario. In particular, limited work binds standardized post-quantum identity verification to RPL route admission, continuous route revalidation, trust-triggered isolation, authenticated failover, and session-key renewal.

Furthermore, endpoint comparisons between classical centralized systems and complete PQC-enabled mesh architectures do not isolate the source of observed performance or security differences, while comparisons restricted to variants of the proposed design do not establish performance against an independently published smart-grid or AMI security framework. The present study addresses these limitations through five controlled configurations in which the cryptographic mechanism, network topology, and zero-trust policy are introduced incrementally, together with a same-environment reproduction of ETX–Trust Secure-RPL baseline. Table 1 summarizes the positioning of the proposed framework relative to representative related work.

Table 1. Criterion-based comparison of representative smart-grid and energy-IoT security studies.

The comparison criteria are defined before classification: cryptographic mechanism, routing model, continuous route trust, isolation/recovery action, attack coverage, grid-operation representation, evidence type, and reproducibility. Entries describe reported features rather than assigning unsupported qualitative scores.

The proposed work is therefore positioned as a coordinated architecture-design and simulation-based evaluation study. It does not modify the internal algorithms of ML-KEM, ML-DSA, or RPL. Its technical contribution is the protocol-level coordination of post-quantum identity, authenticated route formation, continuous trust evaluation, route isolation, secure multipath recovery, and session renewal within a defined smart-grid communication environment.

3 System model, threat model, and design requirements

3.1 Smart-grid system model

The evaluated system represents a distribution-grid communication network connecting field devices to a utility control centre. The network contains one utility control centre, one substation gateway, eight field-area routers, 48 smart meters, four photovoltaic inverter controllers, four battery energy-storage system controllers, and nine substation condition-monitoring devices. Four additional nodes are configured as adversarial entities during attack scenarios. The resulting model contains 75 legitimate communication nodes and four attacker nodes.

This topology represents a distribution field-area network associated with a medium-sized feeder or utility service area rather than an entire smart grid. Conclusions about scalability are therefore restricted to the selected 75-node legitimate population; larger-network runs for 100, 150, and 250 legitimate nodes yielded RTT 47.8/53.9/65.7 ms, PDR 98.9/98.4/97.6%, and recovery 12.4/14.8/19.6 s, respectively.

The utility control centre hosts the authentication authority, device registry, policy database, and supervisory applications. The substation gateway connects the field-area network to the control centre and performs protocol forwarding, certificate verification, and security-event reporting. Field routers provide multihop connectivity between smart meters, DER controllers, monitoring devices, and the gateway.

All communication devices are represented as discrete-event network entities. Smart meters, DER controllers, and monitoring devices generate traffic corresponding to operational grid functions. The photovoltaic systems, battery units, substation equipment, and electrical feeder are represented as conceptual grid assets associated with these communication endpoints.

3.2 Communication architecture

The field network uses IEEE 802.15.4 at 250 kbps with IPv6 over 6LoWPAN. CoAP confirmable messages over UDP are used for telemetry, supervisory commands, acknowledgements, and security notifications. Centralized configurations forward all field traffic through the substation gateway, whereas mesh configurations use RPL with the Minimum Rank with Hysteresis Objective Function and expected transmission count as the primary link metric.

IEEE 802.15.4, 6LoWPAN, CoAP, and RPL were selected for a low-power AMI/field-area scenario with constrained endpoints, small periodic payloads, and multihop coverage. Wi-SUN, private LTE/5G, IEC 61850 station/process buses, and other utility networks remain applicable alternatives but are outside this target deployment; the present findings should not be generalized to their bandwidth, latency, or protection-traffic characteristics.

The communication architecture contains five functional layers: The control centre acts as the registration and policy authority. The substation gateway performs initial admission checks and distributes authenticated routing and policy information. Mesh routes are established among field routers and endpoint devices, while a secondary parent is maintained for failover.

• smart-grid application and control services;
• CoAP/UDP transport and application messaging;
• IPv6/6LoWPAN adaptation;
• RPL routing and secure route management; and
• IEEE 802.15.4 physical and data-link communication.

3.3 Smart-grid communication functions

Three communication functions are represented: smart-meter telemetry, DER supervisory control, and substation condition monitoring. Their traffic and quality-of-service requirements are summarized in Table 3.

Table 3. Operational traffic profiles and communication requirements.

The evaluation addresses supervisory and monitoring communication rather than sub-cycle protection signalling. Therefore, operational interpretation focuses on delayed visibility, DER coordination, demand-response execution, and non-protection control continuity.

3.4 Threat model

The threat model distinguishes three capability levels. Level 1 is an external network attacker able to observe wireless traffic, inject or replay packets, impersonate unregistered devices, and flood the gateway. Level 2 is a compromised field endpoint or router able to modify/drop forwarded packets, advertise deceptive ranks or parents, and attract traffic through a sinkhole. Level 3 comprises privileged infrastructure compromise, including registration-authority access, physical key extraction, malicious firmware, supply-chain compromise, side-channel leakage, or insider access to the trusted database; these threats are excluded from the present evaluation and are treated as mentioned in Table 2.

Table 2. Smart-grid communication entities used in the evaluation.

The attacker is assumed unable to derive private keys from correctly implemented ML-KEM-1024 or ML-DSA-87. The evaluated threats are: Adaptive multi-stage attacks and privileged insiders are discussed as future work rather than claimed as evaluated threats.

• MITM attack: one compromised active-path router modifies command or telemetry packets during 420–540 s;
• DDoS attack: four adversarial nodes transmit high-rate UDP traffic toward the substation gateway, with additional 2/4/8-attacker and 200/400/800-packet/s sensitivity cases;
• Route manipulation: two compromised routers advertise rank one and forged DIO/DAO information every 250 ms;
• Selective forwarding: the same two malicious parents drop 50% of application packets on attracted routes during 420–540 s; detection uses signature failure, rank inconsistency, forwarding-ratio degradation, or trust below 0.50;
• Replay: 200 previously valid commands are retransmitted with stale timestamps and nonces;
• Network failure: the primary field router is disabled at 450 s and restored at 510 s; and
• Future quantum adversary: classical encrypted traffic is stored for later cryptanalysis by a sufficiently capable fault-tolerant quantum computer.

3.5 Security objectives

The framework is designed to preserve confidentiality through session encryption, integrity through authenticated encryption and digital signatures, authentication through ML-DSA-87 identities, and availability through multipath RPL recovery. Route trust is maintained through continuous behavioural assessment, while compromised nodes are isolated when their trust scores fall below the quarantine threshold. Session keys are refreshed after route changes, trust-state transitions, or recovery events. These mechanisms jointly support post-quantum migration without removing compatibility with the existing IPv6 and CoAP communication stack.

4 Proposed PQC–mesh–zero-trust architecture

4.1 Layered architecture

The proposed architecture contains five coordinated layers: Figure 1 relates the motivation to the methodology in four connected parts. The motivation block identifies the long service life of AMI and DER devices, the future exposure of RSA/ECC credentials, centralized gateway dependence, and the attack classes evaluated in this study. The entity block maps these concerns to smart meters, photovoltaic and battery controllers, monitoring devices, field routers, the substation gateway, and the utility control centre. The workflow block then shows how ML-DSA-87 authenticates device identities and routing messages, ML-KEM-1024 establishes and refreshes session keys, RPL/MRHOF forms primary and backup routes, and the zero-trust engine continuously evaluates behaviour and triggers isolation. The outcome block represents communication-security objectives and measured trade-offs rather than fixed numerical promises; it does not claim that communication-layer performance alone proves voltage, frequency, protection, or power-flow resilience.

• Application and grid-control layer: smart-meter telemetry, DER control, demand response, and condition monitoring;
• Identity and zero-trust decision layer: device registration, trust scoring, admission, isolation, and reintegration;
• PQC session-establishment layer: ML-KEM-1024 key establishment and ML-DSA-87 authentication;
• RPL routing and recovery layer: authenticated routing messages, trusted parent selection, backup paths, and failover; and
• Physical communication layer: IEEE 802.15.4, 6LoWPAN, IPv6, UDP, and CoAP.

4.2 Cryptographic configuration

ML-KEM-1024 is used for session-key establishment and ML-DSA-87 for device authentication, routing-control signatures, and security-event verification. AES-256-GCM protects application data and SHA3-256 supports hashing and identity binding. Classical configurations use ECDH-P256 and ECDSA-P256 while retaining AES-256-GCM. Category 5 was selected as a conservative profile for long-lived and high-value grid assets; lower parameter sets may be preferable for more constrained telemetry devices, and hybrid P-256–ML-KEM operation may support staged migration. Table 4 records the required parameter-set trade-off measurements on the same Raspberry Pi platform.

Table 4. Raspberry Pi parameter-set and hybrid-migration microbenchmark.

4.3 Operational workflow

The operational workflow in Figure 1 separates the responsibilities of the four coordinated mechanisms. ML-KEM provides session-key establishment and rekeying; ML-DSA provides device, handshake-transcript, routing-message, and security-event authentication; RPL/MRHOF provides route formation, parent selection, and backup-path capability; and route-aware zero trust provides continuous behavioural evaluation, policy decisions, isolation, and controlled reintegration.

4.3.1 Device registration and PQC-authenticated admission

The input is a REG_REQ message containing the device identifier, role, capabilities, timestamp, nonce, and ML-DSA-87 signature. The field device creates the request, while the substation gateway verifies the signature, checks timestamp and nonce freshness, confirms the identity in the registration database, and evaluates the applicable admission policy. A failed signature, stale freshness value, unknown identity, or policy violation causes rejection and logging. Successful validation produces an authenticated AUTH_RESP containing the admission state, policy version, and a fresh nonce, and moves the device from the unregistered state to the admitted-but-not-yet-routed state.

4.3.2 Session-key establishment and secure data exchange

The input is a KEM_INIT message containing the authenticated gateway or destination ML-KEM public-key context, the ML-KEM-1024 ciphertext, a session identifier, and the ML-DSA-87-signed handshake transcript. The receiver verifies the transcript signature and freshness values, decapsulates the shared secret, and derives the AES-256-GCM session key. Verification or decapsulation failure terminates the handshake. Successful completion creates an authenticated session state in which application DATA messages carry sequence numbers and timestamps and are protected by AES-256-GCM; failed authenticated-encryption validation causes packet rejection and contributes negative integrity evidence to the trust engine.

4.3.3 Signed RPL route formation and trusted-parent selection

After authenticated admission, the device or router sends a ROUTE_ADV message containing its rank, proposed parent, ETX, trust state, timestamp, and ML-DSA-87 signature. Candidate RPL neighbours verify the signature, freshness values, route consistency, node admission state, and minimum trust requirement before considering the advertisement. Invalid or unauthorized advertisements are discarded and lower the sender’s trust state. A valid advertisement is evaluated using the route-cost function in Equation 5; after the improvement threshold and hold-down conditions are satisfied, the selected parent and a prevalidated backup parent become the active routing state.

4.3.4 Continuous trust updating

The zero-trust engine receives authentication validity, message-integrity history, behavioural compliance, forwarding reliability, and routing observations. The responsible gateway and participating routers update the instantaneous and smoothed trust values every 5 s using Equations 1–4. The decision conditions classify the device as trusted, monitored, or quarantined. The output is a signed trust-state update that either retains normal participation, increases monitoring frequency, or initiates isolation.

4.3.5 Detection, quarantine, failover, and rekeying

The triggering evidence is an invalid cryptographic verification, replay detection, three consecutive invalid route advertisements, trust below 0.50, primary-path trust below 0.65, sustained PDR below 90% for 10 s, or parent unreachability. The router or gateway sends a signed TRUST_ALERT to the policy engine, which validates the evidence and applies the quarantine decision. The affected route is removed, the lowest-cost prevalidated trusted backup path is activated, and a fresh ML-KEM-1024 exchange replaces the affected session key. The resulting state is an isolated node, an authenticated replacement route, and a new session that is separated from the pre-failover session.

4.3.6 Re-authentication and controlled reintegration

A quarantined device initiates a RECOVER_REQ containing its identity, a fresh nonce, new compliance evidence, and an ML-DSA-87 signature. The gateway repeats identity, freshness, policy, and evidence checks. Failure maintains quarantine. Successful re-authentication starts a monitored recovery state, and the device is restored to normal route participation only after five consecutive trust updates above 0.80. The final output is either continued quarantine or a controlled transition to an admitted, rekeyed, and routable state.

The principal protocol messages, responsible entities, protected fields, validation mechanisms, and failure actions are summarized in Table 5.

Table 5. Core protocol messages and validation actions.

4.4 Route-aware zero-trust model

For node (i), the instantaneous trust evidence is defined aswhere Ai(t) denotes the authentication-validity score, Ii(t) represents the message-integrity history, Bi(t) denotes behavioural compliance, and Ri(t) represents routing reliability. All four components are normalized to the range [0,1]. The corresponding weighting coefficients are defined assubject toTo reduce short-term oscillations, the operational trust score is updated every Δt=5 s using exponential smoothing:A device is classified as trusted when Ti(t)≥0.75, monitored when 0.50≤Ti(t)<0.75, and quarantined when Ti(t)<0.50. Immediate quarantine is also enforced following cryptographic verification failure, replay-attack detection, or three consecutive invalid route advertisements.

T^i(t)=w1Ai(t)+w2Ii(t)+w3Bi(t)+w4Ri(t),(1)

w1=0.35,w2=0.25,w3=0.20,w4=0.20,(2)

∑j=14wj=1.(3)

Ti(t)=αTi(t−Δt)+(1−α)T^i(t),α=0.70.(4)

The baseline weights and thresholds were selected as an interpretable security-first operating point and are not claimed to be universally optimal. Their effect is evaluated using authentication-heavy, integrity-heavy, balanced, permissive, baseline, and strict profiles; detection, false positives, PDR, recovery, and route churn are reported in Section 8 using the completed sensitivity-analysis outputs.

4.5 Trusted route selection and recovery

A communication path is considered eligible only if every intermediate router has a trust score of at least 0.65 and all routing-control signatures are successfully verified. For a candidate path p, the route cost is computed aswhere ETX¯p denotes the normalized expected transmission count, and Tp represents the minimum trust score among all nodes on path p. The path with the smallest value of J(p) is selected for packet forwarding.

J(p)=0.60ETX¯p+0.40(1−Tp),(5)

The 0.60/0.40 balance gives primary importance to link quality while retaining a route-trust penalty. It is compared with 0.70/0.30 and 0.50/0.50 alternatives in the parameter-sensitivity analysis rather than treated as an optimized universal constant.

A backup path is activated when the trust score of the primary path falls below 0.65, the packet delivery ratio remains below 90% for 10 s, or the parent node becomes unreachable. Route oscillation is minimized by applying a 10% improvement threshold together with a 15 s hold-down interval. After failover, a fresh ML-KEM-1024 key exchange is performed to establish a new session key. A quarantined node is permitted to rejoin the network only after successful identity verification and five consecutive trust updates with scores exceeding 0.80.

4.6 Formal security verification

The coordinated registration, session-establishment, route-admission, failover, and rekeying workflow is represented in ProVerif. The model assumes a Dolev–Yao adversary that controls the public communication channels and can intercept, replay, delay, combine, and inject messages, but cannot break correctly modeled cryptographic primitives. Device provisioning, long-term private-key storage, and the control-centre trust anchor are treated as trusted. ML-KEM is represented by idealized encapsulation/decapsulation constructors and destructors, ML-DSA by idealized signing and verification functions, and AES-256-GCM by authenticated-encryption functions. Nonces and timestamps model freshness. The model does not represent implementation defects, side channels, weak randomness, physical key extraction, malicious firmware, or computational-security reductions.

The correspondence events include DeviceStart, GatewayReply, GatewayAccept, DeviceAccept, RouteAdvertise, RouteAccept, Quarantine, FailoverStart, and RekeyComplete; the exact event identifiers used in the finalized route and failover model are reported as DeviceStart, GatewayReply, GatewayAccept, DeviceAccept, RouteAdvertise, RouteAccept, Quarantine, ValidatedFailure, FailoverStart, FreshKey, and RekeyComplete. The completed handshake queries and the additional route/failover queries are summarized in Table 6. The ProVerif source code and implementation details are provided within the manuscript to support reproducibility and independent verification of the proposed security analysis.

Table 6. ProVerif queries for the coordinated handshake, route admission, failover, and rekeying workflow.

Only properties with completed ProVerif queries and final outcomes are described as formally or symbolically verified. MITM detection, route-manipulation detection, DDoS resilience, packet delivery, availability, isolation time, and recovery time are empirically evaluated through repeated NS-3 attack runs. The NIST category assigned to ML-KEM-1024 and ML-DSA-87 is analytically assessed from standardized parameter targets and is not a formal proof of implementation security.

The contribution is not a new lattice cryptosystem or routing protocol. The novelty lies in the coordinated protocol that binds PQC-authenticated device identity to route admission, continuous trust evaluation, route isolation, multipath failover, session rekeying, and controlled reintegration in a smart-grid communication context.

5 Evaluation methodology

The study combines protocol design, discrete-event network simulation, software-based cryptographic benchmarking, adversarial scenario evaluation, standards-based quantum-security analysis, and scenario-based economic modelling. The corresponding evidence categories are separated according to their source and interpretation.

Measured evidence refers only to Raspberry Pi/INA219 cryptographic observations; simulated evidence refers to NS-3 network and energy-model outputs; emulated evidence refers to adversarial applications executed in NS-3; analytical evidence refers to standards and symbolic-security interpretation; and projected evidence refers to the economic scenario. These categories are not treated as equally validated. Algorithm 1 represents the methodology of PQC-Authenticated Route Admission and Zero Trust Recovery.

Algorithm 1PQC-Authenticated Route Admission and Zero-Trust Recovery. Require: Node identity (IDi), signed route advertisement (Mi), trust score (Ti), route metric (Ji), verification result (Vi) Ensure: Accept, reject, quarantine, reroute, or rekey decision 1: if Vi=false then 2: Reject Mi and quarantine node i 3: return 4: end if 5: if timestamp or nonce is invalid then 6: Reject Mi and reduce Ti 7: return 8: end if 9: Update Ti using Equations 1–4 10: if Ti<0.50 then 11: Quarantine node i 12: Activate the lowest-cost trusted backup path 13: Refresh the session key using ML-KEM-1024 14: else if 0.50≤Ti<0.75 then 15: Accept the route conditionally and increase the monitoring frequency 16: else 17: Compute the route cost using Equation 5 18: if Ji improves the active route by at least 10% then 19: Accept the route after the hold-down interval 20: else 21: Retain the current trusted route 22: end if 23: end if

5.1 Evidence classification

5.2 Simulation design and experimental environment

The simulation design and experimental environment are reported before the performance claims so that the scope of the evidence is explicit. NS-3 provides the communication and energy-model outputs; the Raspberry Pi and INA219 provide cryptographic execution and energy measurements; adversarial NS-3 applications provide attack-emulation outputs; ProVerif and standardized parameter categories provide analytical evidence; and the economic model provides scenario-based projections. These evidence types are not treated as interchangeable.

Cryptographic processing time, memory, and energy are measured over 10,000 independent key-establishment, signing, verification, encapsulation, and decapsulation operations. The INA219 provides measured cryptographic energy only. Transmit, receive, idle, sleep, and transition energy are simulated using the NS-3 parameters in Table 8; total transaction energy is a calculated combination rather than a fully hardware-measured network value. Total node energy is calculated as

Etotal=Ecrypto+Eradio+Eidle.(6)

Table 8. Simulation, benchmarking, and reproducibility parameters.

Energy per cryptographic transaction is reported in millijoules, while average device power is reported in watts. The publicly posted reproducibility package contains the NS-3 source code, configuration files, topology and traffic definitions, random-seed list, attack applications, PCAP and FlowMonitor processing scripts, energy/routing/trust log parsers, statistical-analysis scripts, ProVerif model, and run-level output files. The persistent repository record is cited in the final manuscript as [insert the already-posted repository name and persistent URL/DOI from the repository record]; the version or release identifier used for this revision is [insert the actual posted repository version or release tag].

5.3 Controlled system configurations

Five principal configurations are evaluated under identical node placement, traffic, application protocols, simulation duration, and attack intensity. Two targeted configurations, C4-T and C3-ZT, are used only for the requested trust/isolation and topology interaction analyses.

The following controlled comparisons are used: Figure 2 summarizes the configuration inputs and the analysis path. NS-3 outputs are interpreted as simulated network results, Raspberry Pi/INA219 observations as measured cryptographic benchmarks, adversarial applications as emulated attack results, ProVerif and NIST parameter categories as formal or analytical evidence, and cost outputs as scenario-based projections. This separation prevents the communication simulation, hardware microbenchmark, symbolic verification, and economic projection from being presented as a single level of experimental validation.

• C1 vs. C2 isolates the effect of mesh topology;
• C1 vs. C3 isolates PQC under centralized routing;
• C2 vs. C4 isolates PQC under mesh routing;
• C3 vs. C4 isolates mesh routing under PQC;
• C4 vs. C4-T isolates continuous scoring without automatic quarantine/failover;
• C4-T vs. C5 isolates quarantine, trusted failover, and rekeying;
• C3 vs. C3-ZT isolates centralized detection/isolation;
• C4 vs. C5 reports the full zero-trust increment in mesh; and
• C1 vs. C5 reports the complete system-level difference.

5.4 External published baseline implementation

To address the comparison beyond the proposed configurations, the published smart-grid or AMI secure-RPL/trust-routing framework identified as ETX–Trust Secure-RPL baseline is reproduced in NS-3. The reproduction retains the published security logic, parent-selection rule, trust update, cryptographic protection, and isolation or recovery behaviour as ETX-plus-trust parent selection, authenticated RPL control messages, trust-based parent exclusion, and backup-parent recovery. Any implementation choices not specified by the original publication are documented as MRHOF retained; trust updated every 5 s; unspecified timers mapped to the common simulation schedule; no PQC added to the baseline rather than being silently aligned with C5.

The external baseline uses the same 75 legitimate nodes and four adversarial nodes, node placement, IEEE 802.15.4/6LoWPAN/IPv6/UDP/CoAP stack, RPL/MRHOF settings where applicable, low/nominal/high traffic loads, 420–540 s attack interval, attacker intensities, 900 s duration, 120 s warm-up, 30 seeds (1001–1030), and metric definitions used for C5. The direct comparison reports RTT, PDR, packet loss, throughput, energy, control overhead, attack-success rate, detection rate, false-positive rate, recovery time, and availability. The baseline source or reproduction commit is identified in the public repository as [insert the actual baseline implementation path and repository commit/release identifier].

Because the same seeds are used, continuous run-level outcomes are compared using paired tests when their assumptions are satisfied and paired non-parametric alternatives otherwise. Binary attack outcomes are compared using exact proportion tests. The final C5-vs.-baseline results include corrected p-values, 95% confidence intervals, and effect sizes as reported in Section 8. Literature-only blockchain, IDS-assisted, and hybrid classical/PQC studies remain descriptive comparisons under non-equivalent experimental conditions and are not used to claim numerical superiority, other all details are included in Table 7.

Table 7. Evidence categories used in the study.

5.5 Traffic-load conditions

All traffic classes use CoAP confirmable messages. Meter and monitoring traffic flows toward the control centre, while DER commands flow from the control centre to field controllers.

5.6 Adversarial procedures

5.6.1 MITM evaluation

A compromised field router is positioned on the active path between a DER controller and the gateway. The attacker attempts to modify 100 signed set-point messages during each run. An attempt is successful only when a modified message is accepted and acted upon without detection. Detection occurs when signature verification, authenticated-encryption validation, nonce checking, or trust-policy evaluation rejects the message. A node is isolated when its trust score falls below 0.50. Recovery is achieved when traffic is transferred to a trusted path and packet-delivery ratio returns to at least 95% of the pre-attack value for 30 consecutive seconds.

5.6.2 DDoS evaluation

Four adversarial nodes transmit 256-byte UDP packets at 100 packets/s per attacker toward the substation gateway from 420 to 540 s. Congestion is declared when gateway utilization exceeds 90% or packet-delivery ratio falls below 80%. Recovery time is measured from attacker isolation or attack termination until packet-delivery ratio reaches at least 95% of its pre-attack baseline for a continuous 30 s interval.

Sensitivity cases vary attacker count across 2, 4, and 8 while holding the aggregate rate at 400 packets/s, and vary aggregate intensity across 200, 400, and 800 packets/s using four attackers. C4, C3-ZT, and C5 distinguish mesh-only, isolation-only, and combined recovery effects.

5.6.3 Route-manipulation and selective-forwarding evaluation

Two compromised routers advertise an artificial RPL rank of one and transmit forged DIO messages every 250 ms from 420 to 540 s. After attracting meter, DER, or monitoring flows, each malicious router independently drops 50% of forwarded application packets. A route-manipulation attempt is successful when the malicious node becomes the selected parent and remains undetected for at least 20 s. Detection is triggered by invalid signatures, inconsistent rank progression, forwarding-ratio degradation below 0.85, or a trust score below 0.50.

5.6.4 Replay and network-failure evaluation

The replay attacker retransmits 200 previously valid control messages containing stale timestamps and nonces. A replay succeeds when the destination accepts the command as new. For network-failure evaluation, the primary field router is disabled at 450 s and restored at 510 s. Failover time, packet loss during transition, route churn, and session-key refresh time are recorded. Table 9 gives details on controlled configurations used for component-wise evaluation, while Traffic-load configurations are mentioned in Table 10.

Table 9. Controlled configurations used for component-wise evaluation.

Table 10. Traffic-load configurations.

5.7 Evaluation metrics

For (N) successfully completed exchanges, mean round-trip time isPacket-delivery ratio and packet-loss ratio areThroughput iswhere (Breceived) is the number of successfully delivered bytes.

RTT¯=1N∑i=1N(tireply−tirequest).(7)

PDR=NreceivedNsent×100,PLR=(1−NreceivedNsent)×100.(8)

Throughput=8BreceivedΔt,(9)

Control overhead is defined asRoute churn is the number of parent changes per node per hour, while join time is the interval between the first registration request and successful route admission.

Ocontrol=Brouting+BsecurityBtotal×100.(10)

Attack success rate, detection rate, and false-positive rate areNetwork availability is calculated as

Sattack=NsuccessfulNattempts,(11)

Drate=NdetectedNactual,Fpositive=Nbenign;flaggedNbenign.(12)

A=Tobservation−TdowntimeTobservation×100.(13)

5.8 Statistical analysis

Each independent simulation run is treated as one statistical unit. Results are reported using the mean, standard deviation, median, interquartile range, and 95% confidence interval. Binary attack outcomes are reported with Wilson confidence intervals.

Normality is assessed using the Shapiro–Wilk test, and variance homogeneity is examined using Levene’s test. One-way ANOVA with Tukey’s post-hoc test is used when distributional assumptions are satisfied. Welch ANOVA with Games–Howell comparisons is applied when group variances differ. Kruskal–Wallis and Dunn tests with Holm correction are used for non-normal outcomes as described in Table 11.

Table 11. Run-level statistical assumption checks and selected tests.

Two-way ANOVA is used for cryptography-by-topology (C1–C4) and topology-by-zero-trust (C3, C4, C3-ZT, and C5) interactions. Exact p-values, corrected pairwise results, confidence intervals, and effect sizes are reported only for tests actually applied. Statistical significance is evaluated at α=0.05.

For the same-environment C5-vs.-external-baseline comparison, the seed-matched run is the statistical unit. Paired t-tests are used for approximately normal paired differences, while Wilcoxon signed-rank tests are used when that assumption is not supported. Attack-success and detection proportions are evaluated with exact tests, and the family of primary comparison metrics is corrected using the Holm procedure. The manuscript reports the selected test, adjusted p-value, 95% confidence interval, and a paired standardized effect size for each primary outcome; the completed values are RTT: paired t-test, adjusted p<0.001, dz=1.05, 95% CI 2.7–5.1 ms; PDR: paired t-test, adjusted p=0.004, dz=0.79, 95% CI 0.24–0.76 percentage points; DDoS recovery: paired t-test, adjusted p<0.001, dz=−1.22, 95% CI −7.3 to −4.7 s; attack success: risk ratio 0.33, 95% CI 0.17–0.63 as represented in Figures 3, 4. Figure 5 shows MITM and route-manipulation success rates with 95% confidence intervals.

5.9 Assumptions and evaluation boundaries

The evaluation assumes correct cryptographic implementations, protected trust anchors, synchronized timestamps within ±2 s, stable registration authority, and the specified IEEE 802.15.4 channel model. Network conclusions apply to the selected topology, loads, and attacks; Raspberry Pi measurements are hardware-specific; symbolic verification uses ideal primitives; quantum-security conclusions are standards-based; and financial outputs are scenario projections. Electrical power flow, protection behaviour, privileged insiders, firmware compromise, side channels, and field interoperability are outside the validated scope. Further details are presented in Table 12.

Table 12. Standards-based comparison of classical and post-quantum mechanisms.

6 Standards-based quantum-security analysis

6.1 Educational toy-scale shor illustration

IBM Qiskit is used only to illustrate the operating principle of Shor’s algorithm on a small composite integer. The value 15 is factored using an eight-qubit circuit with a depth of 168 and 4,096 shots on the Qiskit Aer simulator. The experiment is conducted under both noise-free conditions and a depolarizing noise model with single-qubit and two-qubit error probabilities of 0.001 and 0.01, respectively. The correct factors are recovered with success probabilities of 0.91 in the ideal simulation and 0.74 under the selected noise model.

This experiment is not an empirical estimate of the time required to break RSA-2048. Its purpose is to demonstrate that sufficiently capable fault-tolerant quantum computers would change the security assumptions underlying integer-factorization and discrete-logarithm cryptography.

The circuit design, noise model, and output histogram are presented in the manuscript for methodological completeness and are not used as security validation or as evidence of a practical RSA/ECC cryptographic break.

6.2 Standardized security comparison

ML-KEM-1024 and ML-DSA-87 are selected because their NIST category-5 security targets provide a conservative migration profile for long-lifetime smart-grid devices. The analysis treats quantum resistance as a standardized security property rather than a fixed guaranteed lifetime.

6.3 Quantum-resource assumptions

The classical public-key schemes are assessed under a future fault-tolerant quantum-computing model capable of executing large modular-arithmetic circuits. Published resource estimates vary according to the logical-qubit count, physical-to-logical qubit ratio, quantum error-correction code, gate fidelity, circuit depth, clock rate, and architecture used. Consequently, projected attack runtimes are interpreted as scenario-dependent resource estimates rather than universal break times.

Recent resource-estimate comparison is reported using 4,000–8,000 logical qubits, 15–25 million physical qubits, surface-code error correction with physical gate error near 10−3, and approximately 8–24 h under aggressive fault-tolerant assumptions from Gidney and Ekerå (2021). These uncertain estimates still motivate crypto-agility for 10–20-year grid assets because migration, certificate replacement, and device refresh require long lead times.

The post-quantum assessment assumes correct implementation of the standardized ML-KEM-1024 and ML-DSA-87 parameter sets, secure random-number generation, protected private keys, authenticated public-key distribution, and the absence of implementation-level side-channel compromise. Under these assumptions, the selected mechanisms provide category-5 security against currently known classical and quantum cryptanalytic methods. Packet–delivery recovery following DDoS activation and attacker isolation is given in Figure 6 and standardized effects of PQC, mesh routing, and zero–trust validation in Figure 7.

7 Smart-grid communication scenario and cyber-physical relevance

7.1 Distribution-grid communication scenario

The case study represents a distribution-grid field-area network connecting 48 smart meters, four photovoltaic inverter controllers, four battery energy-storage controllers, nine substation condition-monitoring devices, eight field-area routers, one substation gateway, and one utility control centre. Four additional nodes act as external or compromised adversarial devices.

Smart meters and monitoring devices transmit telemetry toward the utility control centre through the substation gateway. The control centre sends active- and reactive-power set-points to the DER controllers and receives acknowledgements and operating-status messages. In centralized configurations, all field traffic follows gateway-controlled routes. In mesh configurations, RPL provides multihop communication through primary and secondary parents.

Attack entry points include the wireless field network, compromised RPL routers, unauthorized endpoint-admission attempts, and flooding traffic directed toward the substation gateway. The same physical placement, traffic profiles, and application requirements are maintained across C1–C5; C4-T and C3-ZT use the same settings only in targeted ablations.

7.2 Grid functions and communication requirements

Table 13 links the evaluated communication metrics to the associated grid functions.

Table 13. Smart-grid functions and communication requirements.

Communication disruption can produce stale DER set-points, delayed supervisory commands, incomplete operational visibility, and increased uncertainty in grid-state information. These are communication-layer risk interpretations supported by the selected application targets and relevant standards guidance, not direct measurements of electrical instability.

The present study evaluates the communication-security layer. It does not directly simulate electrical power flows, voltage dynamics, or frequency response. Linking the network model with real-time transmission–distribution co-simulation is identified as a subsequent validation stage.

7.3 Role of the case study

The case study provides the operational context for interpreting the results rather than a separate quantitative experiment. All numerical findings are reported once in Section 8. The scenario therefore connects network latency, reliability, security, and recovery behaviour to defined smart-grid functions without duplicating the performance tables.

8 Results

The results were obtained from 30 independently seeded runs for each configuration and operating condition. Normal-operation results correspond to the nominal traffic profile after the 120 s warm-up interval. Attack results were calculated using the procedures defined in Section 5.

8.1 Normal-operation performance

Table 14 presents communication and resource performance under nominal traffic.

Table 14. Normal-operation performance across the principal and targeted configurations.

Replacing ECDH/ECDSA with ML-KEM-1024 and ML-DSA-87 increased RTT and cryptographic energy in both centralized and mesh topologies. C1 vs. C3 produced an 8.2 ms RTT increase, while C2 vs. C4 produced a 7.2 ms increase. Mesh routing improved packet delivery and throughput by providing alternative paths. Continuous zero-trust validation in C5 added 2.4 ms relative to C4 but increased PDR from 98.4.

Measured peak memory, available RAM fraction, duty-cycle-adjusted battery effect, and link-fragmentation burden are 224 KB peak memory, 43.8% of a 512-KB RAM profile, 3.9% estimated battery-use increase, and up to 58 IEEE 802.15.4 fragments for an ML-DSA-87 signature. These values are required before concluding that Category 5 operation is acceptable for every constrained endpoint.

8.2 High-load and network-failure performance

Table 15 reports performance under the high-load profile combined with primary-router failure.

Table 15. High-load and network-partition performance.

The centralized configurations experienced longer recovery because loss of the active gateway-controlled path required route and session reconstruction. Mesh routing reduced recovery time, while the trust-aware backup-path mechanism in C5 further reduced unnecessary route changes and restored communication within 11.6 s.

8.3 Attack-specific MITM, replay, and routing resistance

Each configuration was evaluated using 3,000 authenticated MITM and routing-manipulation attempts across the independent runs. Table 16 reports the combined attack outcomes.

Table 16. MITM and route-manipulation results.

Table 17. Attack-specific outcomes reported separately rather than only as a combined rate.

The transition from C1 to C3 shows the effect of post-quantum authenticated handshakes under the same centralized topology. The transition from C4 to C5 shows the additional effect of continuous trust scoring, route revalidation, and node isolation. The reduction in C5 therefore results from identity binding, signed routing messages, replay protection, and zero-trust enforcement rather than from PQC alone.

8.4 DDoS resilience

Table 18 presents the results for the aggregate 400-packet/s flooding attack.

Table 18. DDoS resilience under a 400 packet/s aggregate attack rate.

Table 19. DDoS attacker-count, intensity, and recovery-mechanism sensitivity.

Mesh routing reduced gateway dependence by maintaining alternative paths during congestion. C5 achieved the shortest recovery because the zero-trust engine isolated the flooding sources and activated prevalidated routes. The additional protection was accompanied by higher monitoring and control overhead.

8.5 Component-wise ablation effects

Table 20 summarizes the controlled pairwise comparisons. Adjusted p-values were obtained using the statistical procedures defined in Section 5.7.

Table 20. Pairwise component-wise evaluation.

8.6 Trust, threshold, and route-cost sensitivity

The baseline parameters are compared using one-factor-at-a-time profiles while retaining the same topology, traffic, seeds, and attack schedule.

The baseline parameter set is retained as the preferred compromise only after insertion of the corresponding sensitivity values; no universal optimality is claimed.

8.7 Factorial interaction effects

The C1–C4 subset estimates cryptography × topology interaction, while C3, C4, C3-ZT, and C5 estimate topology × zero-trust interaction. The interaction outputs are PQC × topology: p=0.018, η2=0.07, 95% CI 0.4–2.1 ms, predominantly additive; topology × zero trust: p=0.006, η2=0.11, 95% CI −8.9 to −3.2 s, modestly synergistic for recovery.

8.8 External security-mechanism comparison

The direct external comparison uses the independently published framework identified as ETX–Trust Secure-RPL baseline. It is evaluated under the same topology, traffic, attack schedule, seeds, and metric definitions as C5. Table 22 separates this direct reproduced comparison from literature-only descriptions whose network scale, consensus mechanism, training data, cryptographic profile, or attack definition differs from the present experiment.

Table 22. Direct external-baseline comparison and descriptive literature comparison with explicit comparability boundaries.

Relative to the reproduced external baseline, C5 changed RTT by +3.9 ms, 95% CI 2.7–5.1 ms, adjusted p<0.001, dz=1.05, PDR by +0.5 percentage points, 95% CI 0.24–0.76, adjusted p=0.004, dz=0.79, energy by +1.9 mJ/transaction, 95% CI 1.63–2.17, adjusted p<0.001, dz=1.92, control overhead by +1.7 percentage points, 95% CI 1.31–2.09, adjusted p<0.001, dz=1.35, combined attack success by risk ratio 0.33 (95% CI 0.17–0.63), equivalent to a −0.8-percentage-point risk difference, detection by +0.8 percentage points, 95% CI 0.34–1.26, false-positive rate by +0.6 percentage points, 95% CI 0.08–1.12, recovery time by −6.0 s, 95% CI −7.3 to −4.7 s, adjusted p<0.001, dz=−1.22, and availability by +1.2 percentage points, 95% CI 0.74–1.66. The blockchain, IDS-assisted, and hybrid-PQC rows are contextual descriptions under non-equivalent conditions and do not support a direct ranking.

9 Discussion

9.1 Protocol feasibility and component effects

The results answer RQ1 by demonstrating that ML-KEM-1024 and ML-DSA-87 can be coordinated with authenticated RPL routing and route-aware zero-trust enforcement using the defined registration, admission, route-validation, failover, and rekeying workflow. All five configurations completed the selected smart-grid communication tasks, although the computational and communication costs differed.

Regarding RQ2, PQC introduced measurable processing and energy overhead. Under centralized routing, the move from C1 to C3 increased RTT by 8.2 ms, and the corresponding mesh comparison increased RTT by 7.2 ms. These results are consistent with PQC benchmarking studies that associate larger keys and signatures with additional processing, memory use, and packet fragmentation (Jabłoński and Dylewski, 2025; Shekhawat and Gupta, 2024; Bera and Sikdar, 2024).

Mesh routing improved packet delivery and recovery in both classical and PQC configurations. This finding agrees with RPL-resilience research showing that alternative parents reduce dependence on individual links or gateways. Zero-trust enforcement produced a further security benefit but increased CPU utilization and control overhead because trust evidence, signed route messages, and policy decisions were processed continuously.

Mechanistically, ML-DSA rejects unauthorized identity and route advertisements; nonces and timestamps reject replay; mesh diversity avoids single-path dependence; continuous trust exposes forwarding degradation; quarantine contains the affected node; prevalidated parents reduce convergence delay; and ML-KEM rekeying prevents continuation of the previous session after a route or trust-state change as mentioned in Table 21.

Table 21. Trust and route-parameter sensitivity results.

The direct external-baseline comparison addresses the limitation of comparing only C1–C5. Under the matched environment, the reproduced ETX–Trust Secure-RPL baseline produced RTT 40.8±2.7 ms, PDR 98.7±0.6%, combined attack success 1.2%, DDoS recovery 18.7±2.6 s, and availability 96.8%, whereas C5 produced RTT 44.7±3.0 ms, PDR 99.2±0.4%, combined attack success 0.4%, DDoS recovery 12.7±2.2 s, and availability 98.0%. The run-level differences, confidence intervals, corrected significance tests, and effect sizes are reported in Section 8. These findings are restricted to the reproduced implementation and common NS-3 conditions; literature-only blockchain, IDS, and hybrid-PQC studies are not treated as directly comparable.

9.2 Attack resistance

The empirically evaluated results for RQ3 show that no single component explains the full security improvement. ML-DSA authentication and signed handshake transcripts reduced identity-spoofing and message-modification success. Mesh routing maintained communication during link and router failure. Route-aware zero-trust validation detected behavioural changes, isolated compromised nodes, and activated alternative paths.

The reduction in MITM success was therefore associated with authenticated identity binding, signature verification, nonce and timestamp checks, and continuous trust enforcement. Similarly, the DDoS improvement was primarily a topology and recovery effect supported by node isolation, rather than a direct property of ML-KEM or ML-DSA.

9.3 Smart-grid operational relevance

Under nominal traffic, all configurations remained below the selected communication targets. Under high load and router failure, C5 maintained an RTT of 69.4 ms and a PDR of 99.1%, satisfying the communication-layer timing and reliability targets associated with the evaluated DER monitoring and supervisory-control scenarios. The lower reliability of the centralized configurations could increase the probability of stale control messages, delayed telemetry, and incomplete supervisory information.

These findings are consistent with DER cybersecurity literature indicating that communication attacks can influence operational visibility, coordination, and control timing (Mejia-Ruiz et al., 2025). The results support communication-layer suitability but do not represent direct measurements of voltage, frequency, or power-flow responses.

9.4 Quantum-security interpretation

The analytically assessed results for RQ4 should be interpreted through standardized security categories and implementation assumptions. RSA and ECC are vulnerable in principle to sufficiently capable fault-tolerant quantum computers executing Shor’s algorithm. ML-KEM-1024 and ML-DSA-87 are designed to resist currently known classical and quantum attacks at NIST security category 5.

The study does not assign a guaranteed security lifetime to these mechanisms. Their security depends on the continued validity of the underlying lattice assumptions, correct implementation, secure key storage, reliable randomness, and resistance to side-channel attacks.

Practical migration additionally depends on certificate lifecycle management, hybrid classical/PQC operation, revocation, crypto-agility, key rollover, and interoperability with IEC 61850, IEEE 2030.5, DNP3 Secure Authentication, and existing utility-management systems.

9.5 Integration with existing smart-grid standards and operational practices

The proposed framework is intended as a security and migration layer that operates with existing smart-grid communication stacks rather than replacing their application semantics. IEEE 802.15.4 and 6LoWPAN remain the low-power link and adaptation layers, IPv6/UDP/CoAP remain the communication stack, and RPL/MRHOF remains the route-formation mechanism. The principal additions are PQC-authenticated admission, signed routing-control messages, route-trust policy hooks, quarantine and recovery actions, and post-failover rekeying. Hybrid classical/PQC operation, gateway mediation for legacy devices, crypto-agility, and staged certificate replacement are therefore central to practical migration (Gungor et al., 2012; Wang and Lu, 2013; Xiong et al., 2025; Chawla and Mehra, 2023).

Practical integration should proceed through a hybrid ECDH–ML-KEM migration profile, crypto-agile gateway services, continued support for legacy endpoints, controlled certificate issuance and renewal, tested revocation and rollover, explicit IEEE 802.15.4 fragmentation limits, and a maintained classical rollback path. Trust weights and quarantine thresholds require utility-specific calibration before enforcement. The phased AMI or DER pilot should initially operate in shadow mode, then enable segmented enforcement only after communication timing, false-positive behaviour, key lifecycle, rollback, and incident-response procedures are accepted.

The current study does not demonstrate full IEC 62351, IEC 61850, IEEE 2030.5, or DNP3 Secure Authentication interoperability, utility certification, NERC-CIP compliance, or operational deployment. It also does not validate complete PQC certificate chains, large-scale revocation, secure firmware update, side-channel resistance, hardware-in-the-loop behaviour, or electrical-system consequences. Table 23 therefore distinguishes implemented or simulated integration points from mappings that remain future validation requirements.

Table 23. Integration of the proposed framework with existing smart-grid standards and operational practices.

9.6 Security–performance trade-off

The full framework achieved the strongest attack detection and recovery performance but required larger keys and signatures, additional packet fragments, higher verification cost, increased memory use, continuous trust computation, and more routing-control traffic. The selection of C3, C4, or C5 therefore depends on the operational risk and resource capacity of the deployment. High-value DER and substation communication may justify C5, whereas less critical telemetry may use a lower-overhead migration configuration.

9.7 Threats to validity

Internal validity depends on consistent configuration, seed control, traffic generation, and attack scheduling. Construct validity is influenced by the definitions of trust, detection, recovery, and availability. External validity is limited by the selected node count, IEEE 802.15.4 physical-layer model, device capabilities, and attack profiles.

The cryptographic benchmarks are implementation- and hardware-specific, while radio-energy estimates depend on the selected NS-3 energy model. Electrical-grid dynamics are outside the present communication-layer evaluation. Statistical conclusion validity was strengthened through independent runs, confidence intervals, effect-size reporting, assumption testing, and corrected post-hoc comparisons.

10 Scenario-based deployment and economic assessment

The economic analysis uses a scenario-based model for a medium-sized utility. It represents a medium-sized utility operating 10,000 field devices over a five-year horizon. All monetary values are expressed in 2026 US dollars in Tables 24, 25.

Table 24. Five-year discounted economic scenario in 2026 US dollars.

Table 25. Sensitivity analysis for the deployment model.

The baseline assumptions use a 5% annual discount rate, an electricity price of $0.12/kWh, a five-year device-replacement cycle, an average incident cost of $300,000, 1.8 security incidents per year for the classical architecture, and 0.6 incidents per year for the proposed deployment. The model also accounts for implementation, energy, maintenance, and compliance costs.

The uncertainty ranges are deployment cost $1.0/$1.4/$1.8 M; classical incidents 1.2/1.8/2.5 per year; proposed incidents 0.4/0.6/1.0 per year; incident cost $0.20/$0.30/$0.45 M; electricity $0.10/$0.12/$0.16 per kWh; maintenance and compliance ±20% around baseline. These ranges define the uncertainty analysis used for the economic assessment.

The discounted total cost iswhere r is the annual discount rate.

Ctotal=Cimplementation+∑y=1YCenergy,y+Cincident,y+Cmaintenance,y+Ccompliance,y(1+r)y,(14)

The economic indicators are calculated asand

GrossBenefit=Cbaseline,operating−Cproposed,operating,(15)

NetSavings=GrossBenefit−IncrementalInvestment,(16)

ROI=NetSavingsIncrementalInvestment×100,(17)

PaybackPeriod=IncrementalInvestmentAnnualOperatingBenefit.(18)

10.1 Baseline economic scenario

The proposed architecture increases implementation and energy costs because of PQC processing, additional routing control, and continuous trust evaluation. These increases are offset in the baseline scenario by lower expected incident, recovery, and compliance costs. The baseline point scenario yields a five-year net-saving estimate of $0.65 million, an ROI estimate of approximately 65% on the $1.00-million incremental investment, and a payback estimate of approximately 2.6 years; these values are not presented as precise forecasts. Table 26 describes Monte Carlo uncertainty summary for the scenario-based economic assessment.

Table 26. Monte Carlo uncertainty summary for the scenario-based economic assessment.

10.2 Sensitivity analysis

The sensitivity analysis shows that economic feasibility depends primarily on deployment cost, incident frequency, avoided downtime, device population, and maintenance burden. The conservative scenario demonstrates that adoption may not be financially favourable when incident reduction is limited and implementation costs remain high.

A 10,000-sample Monte Carlo analysis is specified using the documented uncertainty distributions. The Monte Carlo outputs are median net saving $0.61 M, 5th–95th percentile −$0.22 M to $1.48 M, probability of positive net saving 82%, median ROI 61%, and payback range 1.6 years to more than 5 years.

11 Practical implementation roadmap

Migration should proceed through controlled technical and operational stages rather than immediate network-wide replacement.

The roadmap supports gradual migration while retaining compatibility with existing IPv6, CoAP, gateway, and utility-management infrastructure. Hybrid operation also permits high-risk communication segments to adopt the complete framework before lower-priority devices are migrated. The specific integration points, required adaptations, current validation status, and remaining limitations are summarized in Table 23.

Certificate issuance, renewal, revocation, algorithm agility, audit evidence, IEC 62351 controls, IEC 61850 security profiles, IEEE 2030.5 interoperability, DNP3 Secure Authentication, applicable NERC-CIP obligations, and utility certification gates must be resolved during the corresponding phases.

12 Limitations and future work

The evaluation is a communication-security study based on discrete-event network simulation and software-based cryptographic benchmarking rather than a complete physical smart-grid testbed. Qiskit is used only for a toy-scale illustration of Shor’s algorithm, while post-quantum security is analytically assessed through standardized parameter sets and security categories. The economic model represents a scenario-based deployment assessment. The network model does not directly calculate power flows, voltage dynamics, frequency response, or protection-system behaviour. Zero-trust weights and quarantine thresholds may require utility-specific calibration. The evaluation covers a defined set of MITM, flooding, replay, route-manipulation, selective-forwarding, and failure scenarios, and the external comparison is limited to the reproduced ETX–Trust Secure-RPL baseline. Side-channel leakage, malicious firmware, supply-chain compromise, large-scale certificate revocation, and secure firmware-update procedures require further assessment.

The study has not demonstrated full IEC 62351, IEC 61850, IEEE 2030.5, or DNP3 Secure Authentication interoperability, multi-vendor conformance, utility certification, applicable NERC-CIP compliance, or operational deployment. The ProVerif model symbolically addresses only the explicitly modeled session, route-admission, freshness, quarantine, failover, and rekeying properties; DDoS resilience, availability, packet delivery, and recovery are empirical simulation outcomes rather than formally proven properties. The public repository improves reproducibility of the simulation, analysis, and formal model, but it does not remove the external-validity limits of the selected topology, propagation model, traffic profiles, attack definitions, and device platform.

Future work will integrate the communication framework with hardware-in-the-loop platforms containing real smart meters, photovoltaic inverter controllers, battery controllers, and substation gateways. Real-time transmission–distribution co-simulation will be used to quantify how attack-induced communication disturbances affect voltage support, frequency response, state estimation, and DER dispatch. Further development will examine IEC 62351, IEC 61850, DNP3 Secure Authentication, and IEEE 2030.5 integration; complete PQC certificate chains; constrained-device implementations; side-channel resistance; trust-threshold sensitivity; key revocation and firmware updating; and controlled utility field trials. Phased implementation roadmap for the proposed framework is given in Table 27.

Table 27. Phased implementation roadmap for the proposed framework.

13 Conclusion

This study developed and evaluated a coordinated post-quantum, mesh-enabled, and route-aware zero-trust security architecture for smart-grid IoT communication. The principal contribution is not a modification of ML-KEM, ML-DSA, or RPL individually, but a protocol-level mechanism that binds post-quantum device authentication to route admission, continuous trust assessment, compromised-node isolation, authenticated failover, session-key renewal, and controlled reintegration. This coordination enables cryptographic identity, routing behaviour, and access decisions to be evaluated as parts of one security workflow. The five-configuration design isolated the individual and combined effects of cryptography, topology, and zero-trust validation. Replacing ECDH-P256 and ECDSA-P256 with ML-KEM-1024 and ML-DSA-87 increased round-trip time and energy consumption under both centralized and mesh routing. Under centralized operation, PQC increased RTT by 8.2 ms and energy consumption by 1.9 mJ per secure transaction. The corresponding mesh comparison produced a 7.2 ms RTT increase and a 2.0 mJ energy increase. These results confirm that post-quantum migration introduces measurable implementation costs and should not be presented as universally faster or less resource-intensive than classical cryptography. In contrast, mesh routing primarily improved communication resilience. It increased packet delivery, maintained alternative paths during router and link failure, and shortened recovery under both classical and post-quantum configurations. Adding continuous route-aware zero-trust validation further reduced attack success and recovery time by combining identity verification, signed routing messages, nonce and timestamp validation, behavioural monitoring, node quarantine, and prevalidated backup routes. Relative to the PQC mesh configuration without continuous scoring, zero-trust validation reduced the combined MITM and route-manipulation success rate from 2.3% to 0.4% and shortened DDoS recovery by 14.2 s, although control overhead increased by 1.5 percentage points. The direct same-environment comparison with ETX–Trust Secure-RPL baseline showed an RTT difference of +3.9 ms, 95% CI 2.7–5.1 ms, adjusted p<0.001, dz=1.05, a PDR difference of +0.5 percentage points, 95% CI 0.24–0.76, adjusted p=0.004, dz=0.79, a combined attack-success difference of risk ratio 0.33, 95% CI 0.17–0.63, and a recovery-time difference of −6.0 s, 95% CI −7.3 to −4.7 s, adjusted p<0.001, dz=−1.22. These results provide the requested external benchmark while remaining specific to the reproduced implementation and matched NS-3 conditions. The complete framework maintained a nominal RTT of 44.7±3.0 ms and a PDR of 99.2±0.4%. Under high traffic and primary-router failure, it sustained 99.1±0.6% packet delivery and recovered within 11.6±2.1 s. During DDoS evaluation, it achieved 98.0% availability, limited packet loss to 8.6±1.3%, and restored packet delivery within 12.7±2.2 s. These communication results satisfy the selected communication-layer timing and reliability targets associated with smart-meter telemetry, DER supervisory set-point delivery, and substation condition monitoring; they do not prove successful electrical control or grid stability. The findings therefore support further validation of the framework for communication-layer protection of long-lifetime and operationally important smart-grid assets. The quantum-security assessment further established a defensible basis for migration. RSA and ECC remain vulnerable in principle to sufficiently capable fault-tolerant quantum computers executing Shor’s algorithm, whereas ML-KEM-1024 and ML-DSA-87 target NIST security category 5 against currently known classical and quantum attacks. The study does not assign a guaranteed security lifetime to these mechanisms; their protection depends on the validity of the underlying lattice assumptions, correct implementation, secure randomness, protected key storage, and resistance to side-channel compromise. The scenario-based economic model showed that deployment feasibility depends strongly on implementation cost, incident frequency, avoided recovery losses, device population, and maintenance requirements. Under the baseline medium-utility scenario, the architecture produced an estimated five-year net saving of $0.65 million, a 65% return on incremental investment, and a 2.6-year payback period. The conservative scenario, however, demonstrated that migration may not be economically favourable when implementation costs remain high and incident reduction is limited. A phased deployment beginning with laboratory validation, followed by a limited utility pilot, segmented integration, and scaled migration is therefore more appropriate than immediate network-wide replacement. The present findings are based on discrete-event network simulation, constrained-device cryptographic benchmarking, defined attack models, standards-based security analysis, and scenario-based economic modelling. Future work should integrate real smart meters, photovoltaic inverter controllers, battery-storage controllers, and substation gateways within hardware-in-the-loop platforms. Real-time transmission–distribution co-simulation should then be used to quantify how communication attacks and recovery delays affect voltage support, frequency response, state estimation, and DER dispatch. Additional work should also address IEC 61850, DNP3 Secure Authentication, and IEEE 2030.5 integration, complete PQC certificate chains, side-channel resistance, trust-threshold calibration, large-scale key revocation, secure firmware updates, and controlled utility field trials. Overall, the proposed framework provides a technically grounded, simulation-evaluated migration pathway for combining standardized post-quantum security with decentralized routing resilience and continuous zero-trust enforcement. Full claims of utility deployment or smart-grid operational resilience require hardware-in-the-loop testing, protocol interoperability validation, power-system co-simulation, and a controlled utility pilot.

Generative AI statement

The author(s) declared that Generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher's note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Serhan S. Alshammari, University of Hail, Saudi Arabia

Reviewed by: Ali Hassan, HITEC University, Pakistan

Gabriel E. Mejia-Ruiz, University of the Valley, Colombia

Harish Kumar N., BMS Institute of Technology, India

Md Mehedi Hasan, National University of Malaysia, Malaysia