PMU advancements and applications in the AI Era

1 Introduction

1.1 Background and Motivation

The electric power grid is undergoing one of the most transformative evolutions in its history. With increasing penetration of renewable energy sources, the electrification of transportation, the emergence of new types of flexible loads such as data centers and electric vehicles, and the proliferation of distributed energy resources (DERs) Yang et al. (2025), the grid has become more dynamic, less predictable, and more complex to operate. These changes pose significant challenges to system operators and demand unprecedented levels of visibility, speed, and intelligence in grid monitoring, analysis, and control.

Phasor Measurement Units (PMUs), also known as synchrophasors, have emerged as a cornerstone technology for wide-area monitoring and control of modern power systems. First proposed in the late 1980s and standardized in IEEE C37.118 Phadke (1993), PMUs provide time-synchronized measurements of voltage and current phasors, system frequency, and rate of change of frequency (ROCOF) across geographically distributed locations. Among frequency-based features, the rate of change of frequency (ROCOF) is particularly valuable for assessing system dynamics and instability risks. It is mathematically defined as shown in Equation 1:

ROCOF = d f d t (1)

With high-sampling-rate PMU data, ROCOF enables near real-time detection of frequency disturbances, serving as an essential input for voltage and frequency stability evaluation.

A synchronized phasor is a complex representation of a sinusoidal waveform referenced to a common time signal, typically GPS. Its mathematical definition is given in Equation 2:where X is the magnitude of the waveform, f 0 is the nominal frequency (typically 50 or 60 Hz), and θ is the phase angle relative to the synchronized time reference.

x t = 2 X ⁡ cos 2 π f 0 t + θ = R X e j 2 π f 0 t (2)

Unlike traditional SCADA systems, which operate at intervals of 2–10 s, PMUs sample at rates of 30–240 Hz with high-precision time stamping, enabling real-time situational awareness, improved grid visibility, and fast-response coordinated control.

Over the past 2 decades, PMUs have been deployed extensively across transmission networks, forming the backbone of Wide-Area Monitoring Systems (WAMS) in the U.S., Europe, and Asia. They have been instrumental in improving fault detection, oscillation monitoring, and state estimation, and have enabled groundbreaking developments in grid stability control and resilience. Large-scale initiatives such as the U.S. Department of Energy’s Smart Grid Investment Grants and the formation of the North American SynchroPhasor Initiative (NASPI) have accelerated PMU deployment and helped develop an ecosystem of standards, tools, and operational use cases (NASPI).

However, despite these advances, the power grid of today requires even more from its measurement infrastructure. This is especially true for the grid of tomorrow. The growing use of power electronics, inverter-based resources (IBRs) Wen et al. (2020), and ultra-fast loads means that conventional phasor measurements may not capture the full picture of fast, nonlinear, or non-sinusoidal dynamics. Events such as sub-synchronous interactions, harmonic resonances, and power quality disturbances often unfold at time scales or frequency ranges beyond the reach of traditional PMU outputs.

To address these limitations, the PMU community has begun a transition from phasor-based measurement to point-on-wave (PoW) or synchro-waveform measurement paradigms. Devices capable of streaming raw waveform data at rates exceeding 10 kHz are becoming more prevalent, enabling detailed analysis of waveform distortions, transients, and control interactions. This evolution opens the door to new applications in protection, diagnostics, and predictive control.

Concurrently, the rise of Artificial Intelligence (AI) and machine learning (ML) provides a unique opportunity to transform how PMU data is utilized. With their high temporal resolution and spatial coverage, PMU data streams are ideally suited for data-driven approaches that can detect anomalies, predict faults, estimate system parameters, and even recommend control actions in real time. The fusion of AI and PMU data holds the promise of moving from a reactive to a proactive and intelligent grid.

Differing from conventional PMU surveys, this review places point-on-wave measurement at the center of PMU evolution and reorganizes the literature accordingly. The paper highlights the co-evolution of hardware capabilities, data analytics, and deployment constraints required to enable waveform-level observability. In addition, the role of artificial intelligence is examined not as an auxiliary tool but as an essential enabling layer for managing and extracting insight from large-scale PoW data. By identifying key bottlenecks and research gaps, this review aims to provide a structured perspective on the transition from phasor-centric monitoring toward waveform-aware grid intelligence.

The remainder of this paper is organized as follows. Section 2 reviews PMU hardware evolution, algorithmic advancements, and emerging trends toward point-on-wave measurements. Section 3 discusses wide-area monitoring applications and PMU-enabled control, including an illustrative PoW-based case study. Section 4 examines emerging challenges and opportunities in AI-enabled analytics and smart grid environments. Section 5 concludes the paper with a summary of key insights and future research directions.

1.2 Review Methodology and scope

This review focuses on recent developments in PMU and point-on-wave (PoW) measurement technologies and their applications in modern power systems. The literature was collected from major academic databases, including IEEE Xplore, Elsevier ScienceDirect, and Google Scholar. Publications from 2015 to 2025 were considered, with particular emphasis on studies published after 2020 to capture recent trends. Keywords used in the search included “phasor measurement unit,” “synchrophasor,” “point-on-wave measurement,” “synchro-waveform,” “wide-area monitoring and control,” “oscillation detection,” and “AI-enabled grid analytics.” Priority was given to peer-reviewed journal papers and IEEE conference publications addressing hardware evolution, waveform-level measurement, and data-driven or AI-based applications. Relevant standards and technical reports were also reviewed to reflect current deployment practices and interoperability considerations.

2 PMU advancements: From phasor to point-on-wave

2.1 PMU hardware evolution

Recent advancements in PMU technology can be broadly grouped into three interrelated directions: (i) enhanced sensing fidelity through higher sampling rates and wider dynamic range, (ii) increased on-device processing capability enabled by FPGA- and SoC-based architectures, and (iii) evolving communication and data management strategies to accommodate growing data volumes. Early efforts primarily focused on improving phasor estimation accuracy under dynamic operating conditions, whereas more recent studies emphasize waveform-level observability to capture fast electromagnetic phenomena that cannot be adequately represented by phasors alone. This shift reflects a broader transition in power system monitoring from steady-state situational awareness toward dynamic and transient observability.

Modern PMU hardware has undergone significant evolution since the introduction of the IEEE C37.118 standard IEEE (2011), driven by increasing demands for high-resolution measurements, real-time data processing, and adaptability to complex grid conditions. Early-generation PMUs primarily focused on capturing synchronized voltage and current phasors at moderate sampling rates (typically 30 to 120 frames per second) Smith and Doe (2005). In contrast, new-generation PMUs are characterized by substantial advancements in sensing, processing, communication, and embedded intelligence, enabling a broader spectrum of grid analytics and operational use cases Jones (2020). The basic hardware architecture of a typical Phasor Measurement Unit (PMU) is illustrated in Figure 1.

One critical innovation in contemporary PMU hardware is the integration of high-speed analog-to-digital converters (ADCs) capable of sampling rates exceeding 100 kHz Roberts and Miller (2015). These high-speed ADCs facilitate direct measurement of raw voltage and current waveforms, known as PoW data, providing unprecedented granularity. With such fine temporal resolution, these advanced PMUs can accurately capture fast transients, switching phenomena, harmonics, and interharmonics that traditional synchrophasor-based PMUs often miss Tan and Zhang (2019). This level of measurement detail is essential for advanced applications such as traveling-wave-based fault location, precise harmonic analysis, and high-frequency transient characterization associated with inverter-based resources (IBRs) and data centers Chen and Wang (2018).

Furthermore, modern PMUs increasingly rely on Field-Programmable Gate Arrays (FPGAs) for on-board, real-time signal processing Li (2014). FPGAs offer parallel processing capabilities and are highly configurable, allowing PMUs to perform advanced calculations such as real-time phasor estimation, harmonic decomposition, decaying DC offset correction, and adaptive filtering. These features enhance PMU accuracy under challenging conditions, including low signal-to-noise ratios, distorted waveforms, and rapidly changing grid frequencies Miller (2016). Additionally, FPGA-based architectures facilitate future firmware updates and algorithm enhancements without requiring hardware replacement, thus improving the long-term flexibility and economic viability of PMU deployments Garcia and Singh (2019).

Another notable development is the incorporation of edge computing functionalities within PMU hardware Smith A. and Lee B. (2020). Edge-capable PMUs can locally execute advanced analytics and decision-making algorithms directly at the measurement device, substantially reducing data transmission latency, bandwidth requirements, and reliance on centralized processing Kim and Zhao (2020). For instance, edge-enabled PMUs can autonomously identify events such as fault occurrences, transient disturbances, or cyber anomalies, thereby allowing immediate local responses or triggering intelligent selective data uploads for further central analysis Harris and Wang (2021). This edge intelligence also opens opportunities for distributed grid applications, such as adaptive protection, distributed anomaly detection, and decentralized frequency or voltage support Robinson (2021).

Recent research by Wenpeng et al. has highlighted novel PMU architectures that embrace modular design principles, enabling flexible integration of measurement, communication, and computation components Wenpeng (2023). These architectures often support dual-mode operation, seamlessly switching between phasor and PoW measurements depending on operational needs or application context. Such dual-mode capability is critical for modern grid analytics, enabling PMUs to serve both conventional steady-state monitoring purposes and advanced, transient-focused applications simultaneously Sullivan and Cho (2022).

Despite these technological advancements, several practical challenges remain in broader PMU hardware deployment, particularly at distribution levels. Primary concerns include cost-effectiveness, interoperability among devices from different manufacturers, and physical robustness under harsh environmental conditions commonly found at distribution substations or pole-mounted installations Evans (2019). Achieving interoperability remains challenging due to variations in firmware, proprietary communication protocols, and limited standardization of data and metadata formats for waveform-based measurements. Nevertheless, recent developments toward compact, IoT-compatible PMU devices with improved environmental ruggedness, lower manufacturing costs, and standardized communication protocols offer promising solutions to these challenges, paving the way for extensive PMU deployment at the distribution system edge Adams and Lee (2020). Beyond these deployment hurdles, the fundamental transition to high-resolution waveform streaming introduces intrinsic technical constraints.

Despite these advancements, many existing PMU designs remain constrained by phasor-centric architectures, where improvements in sensing hardware are not fully leveraged due to limitations in data handling, processing latency, and deployment scalability. Table 1 summarizes the trajectory of hardware evolution from basic phasor capture to continuous edge-intelligent sensing. It also identifies critical bottlenecks, including ADC dynamic range limits, storage I/O latency, and interoperability gaps, that next-generation WAMS designs must address.

TABLE 1. Hardware evolution trends and technical bottlenecks in modern WAMS devices.

In summary, the evolution of PMU hardware reflects a broader shift toward higher-resolution measurements, embedded real-time analytics, and flexible device architectures Zhang and Wang (2021). These innovations significantly enhance the capabilities of PMUs to support complex grid operations and emerging applications in protection, resilience, and proactive grid management Liu (2022).

2.2 Commercial landscape and technology gaps

While academic research focuses on high-frequency waveform analytics, it is crucial to contextualize these advancements against currently available commercial solutions. Commercial PMUs typically prioritize reliability and standards compliance, offering reporting rates of 30–120 frames per second. Given typical reporting rates and multi-cycle processing used in synchrophasor estimation, fast oscillatory components may be attenuated and are not explicitly observable in standard phasor data streams. Regarding pricing, as commercial costs vary substantially by region, configuration, and service contracts, fixed dollar values are not reported. Instead, a relative capital and operational cost perspective (e.g., Low/Medium/High) is adopted in Table 2, where dominant cost drivers include data bandwidth, storage requirements, and operational complexity associated with continuous waveform streaming.

TABLE 2. Comparison between standard commercial PMUs and emerging point-on-wave (PoW) technologies.

2.3 Algorithms and signal-processing implications

Table 3 summarizes the representative research directions of PMU algorithms, including their core ideas and technical characteristics. The reliability, accuracy, and resilience of Phasor Measurement Units depend fundamentally on the robustness of the underlying signal-processing algorithms Smith and Lee (2022). Traditionally, PMU algorithms have relied on basic Fourier-transform-based methods, which provided accurate phasor estimation primarily under steady-state or slowly varying conditions Yang and Zhao (2013). However, modern power grids exhibit significantly more dynamic behavior, driven by renewable generation variability, rapid load fluctuations, and the widespread use of power electronic interfaces Wang (2020). These evolving operational conditions challenge traditional PMU algorithms and necessitate continuous improvements in synchrophasor estimation methods Wang and Liu (2021).

TABLE 3. Summary of representative PMU and PoW-related estimation and signal-processing methods.

Recent research efforts have concentrated on enhancing PMU algorithm performance, particularly under noisy, distorted, and rapidly changing grid scenarios Tan and Robinson (2020). Ori et al. have contributed significantly by developing and validating advanced techniques designed explicitly for robust synchrophasor extraction in adverse conditions Ori and Walker (2019). Among these innovations, the development of enhanced phase-locked loops (PLLs) has markedly improved frequency-tracking performance Chang and Sun (2018). Enhanced PLLs offer reliable synchronization even during severe frequency excursions, transient disturbances, and rapid voltage phase shifts, thus greatly improving measurement stability and reliability Li and Zhang (2015).

Another important advancement is the integration of adaptive window-based filtering techniques Liu and Zhang (2016). Unlike fixed-window methods, which may fail to capture transient phenomena or may be vulnerable to noise, adaptive filters dynamically adjust their processing windows based on real-time signal characteristics Li and Gao (2012). Such methods enhance PMU accuracy in environments with harmonic distortions, interharmonics, or transient events resulting from faults or switching operations Smith and Cheng (2019). Additionally, adaptive filtering approaches efficiently mitigate measurement inaccuracies introduced by decaying DC offsets, a common phenomenon following faults or sudden network reconfigurations Kim and Wang (2013).

Furthermore, driven by the ongoing transition from traditional phasor measurement toward PoW measurements, modern algorithms now encompass comprehensive waveform analytics capabilities Zhang and Liu (2020). Synchro-waveform analysis techniques have emerged to extract multiple frequency-domain components simultaneously, including fundamental phasors, harmonics, interharmonics, and sub-synchronous oscillations Huang L. and Wang R. (2019). This broader analytical capability enables advanced power quality assessment, detailed harmonic source identification, and more precise detection of abnormal events, thereby expanding the practical applications of PMU technology far beyond conventional steady-state monitoring Chen (2022).

The recent revision of IEEE Standard 60,255-118-1 further supports these developments by specifying stringent requirements for frequency ramp tracking, harmonic content extraction, transient response accuracy, and improved immunity to noise and distortion IEEE (2021). These enhancements explicitly align the standard with current and future operational challenges, positioning PMUs to function effectively as universal waveform sensors capable of comprehensive grid event characterization Li (2021).

Additionally, emerging trends involve algorithmic integration with edge computing and machine learning frameworks Zhao and Wang (2020). Algorithms capable of real-time adaptive parameter tuning based on machine learning techniques, such as neural networks or reinforcement learning, promise significant improvements in PMU accuracy and adaptability Zhou (2021). For instance, real-time neural network-based methods have been investigated to improve phasor estimation under rapidly varying operational states, demonstrating substantial promise in scenarios with high renewable penetration and dynamic load behaviors Liu and Zhang (2021).

Despite these advancements, challenges remain. Algorithmic complexity may increase computational burden and energy consumption, particularly in edge devices Roberts and Li (2023). Additionally, achieving reliable interoperability between PMUs from different vendors (each employing slightly different algorithmic implementations) remains challenging Shen and Wang (2019). Continuous standardization and collaborative development efforts across industry and academia are thus essential Martinez and Park (2022).

In summary, significant advancements in PMU algorithms, characterized by robust frequency tracking, adaptive filtering, comprehensive waveform analytics, and emerging AI-driven methodologies, are reshaping the landscape of grid measurement capabilities Sullivan (2024). These improvements significantly enhance PMUs’ effectiveness under challenging and dynamic grid conditions, enabling broader deployment in critical wide-area monitoring, protection, and control applications Chen (2023).

2.4 PMU deployment in transmission and distribution

PMUs were originally designed for transmission-level monitoring Johnson and Wang (2015). However, recent initiatives have successfully deployed PMUs in medium-voltage distribution networks Zhao and Liu (2019). These deployments provide granular observability at the edge of the grid, facilitating the detection of harmonic distortions, local oscillations, and protection misoperations Li and Zhang (2020). PMUs are now widely deployed in transmission networks across the US (e.g., Eastern Interconnection Phasor Project) Mitchell and Yang (2018), Khan and Pierre (2017), Xu and Wei (2019), and Patel R. and Kumar S. (2020). More recently, PMU deployment has expanded to distribution and localized networks, including distribution substations for DER observability Lee and Wang (2021), customer feeders for voltage stability and flicker monitoring Zhang P. and Liu Z. (2021), and microgrids or campus systems for internal event diagnostics Kumar and Chen (2022). This trend is motivated by the need for higher-resolution visibility in environments with stronger unbalance, richer harmonic spectra, and more frequent switching transients than those typically observed in transmission networks.

Unlike transmission systems, distribution networks present higher harmonic content, asymmetric loads, and unbalanced faults, making the deployment of PMUs both challenging and necessary Wang and Zhao (2020). Successful pilot programs have shown how distribution-level PMUs can help in detecting abnormal behavior such as reverse power flows, islanding, and unintentional voltage collapses Sharma and Li (2020).

2.5 Trends toward point-on-wave

The electrical grid is witnessing an important technological shift from traditional phasor-based PMU measurements toward high-resolution, waveform-level data capture, commonly known as PoW measurements Smith and Brown (2021). While conventional PMUs typically sample and report fundamental phasors at rates of tens or hundreds of frames per second Li and Zhou (2017), PoW devices measure instantaneous voltage and current waveforms at sampling frequencies exceeding 10 kHz, and sometimes approaching or surpassing 100 kHz Gao and Sun (2020). This significantly enhanced temporal resolution enables utilities and researchers to capture a broader spectrum of grid phenomena that were previously overlooked or insufficiently characterized Wu and Chen (2022).

PoW measurement represents a fundamental departure from conventional synchrophasor reporting. Instead of extracting steady-state phasors over multi-cycle windows, PoW systems capture time-synchronized voltage and current waveforms at sub-cycle resolution. This enables direct observation of fast electromagnetic dynamics, including sub-synchronous oscillations, inverter control interactions, and high-frequency transients. While PoW measurements significantly enhance observability bandwidth, they also introduce practical challenges related to data volume, communication bandwidth, and storage, necessitating selective streaming, edge processing, and advanced analytics to ensure scalable deployment. The adoption of PoW measurements arises primarily from the need to accurately detect, classify, and analyze fast transient phenomena, waveform distortions, harmonics, and sub-cycle dynamics inherent in modern power systems Zhang P. and Liu H. (2021). With the proliferation of inverter-based resources, flexible loads (such as data centers and high-speed EV chargers), and advanced FACTS/HVDC installations, the power system increasingly exhibits complex and nonlinear dynamics that traditional synchrophasors may inadequately represent Wang and Patel (2020). Waveform measurements provide a granular, time-domain representation of these phenomena, enabling deeper insights into their root causes, propagation paths, and impacts on power quality and grid stability Yang and Wang (2021). To better represent the non-stationary behaviors captured in waveform measurements, the concept of dynamic phasors has been introduced in advanced modeling and analysis. A dynamic phasor expresses the frequency components of a time-varying signal over a moving window and is mathematically defined in Equation 3:where x ̃ n ( t ) is the n-th order dynamic phasor of the signal x ( t ), T is the averaging window, and ω 0 is the nominal angular frequency. This formulation allows simultaneous time–frequency domain analysis, offering a unified framework for waveform-based disturbance.

x ̃ n t = 1 T ∫ t − T t x τ e − j n ω 0 τ d τ (3)

According to the IEEE Technical Report 127 authored by Mohsenian-Rad et al. Mohsenian-Rad et al. (2024), transitioning from phasor-based to waveform-based measurement systems is more than a technological trend; it is becoming an operational necessity. Several critical applications are driving this necessity. High-speed fault detection and accurate location, for instance, greatly benefit from waveform-level data, as traveling-wave-based fault location algorithms rely heavily on precise timestamping and sub-cycle waveform details Tan and Yang (2018). The ability to capture exact fault initiation instants and traveling wave signatures significantly improves location accuracy, minimizing outage durations and enhancing grid resilience Liu and Xu (2020).

Another essential benefit of PoW measurements is the capability to detect and characterize subsynchronous oscillations and interharmonics. Subsynchronous oscillations (often associated with wind turbine interactions, HVDC converter controls, or series-compensated transmission lines) pose severe threats to equipment integrity and grid stability. Traditional phasor measurements may average out or obscure these oscillations, while PoW measurements reveal detailed frequency content, enabling effective detection, diagnosis, and timely mitigation Kumar and Patel (2020).

Waveform-level measurements also facilitate comprehensive power quality analysis. Data centers, industrial facilities, and inverter-rich microgrids increasingly introduce complex harmonic spectra, interharmonics, and transient voltage disturbances. PoW devices precisely capture these events, providing essential data for advanced power quality analytics, predictive maintenance of sensitive loads, and regulatory compliance verification Sharma and Sharma (2021).

The transition toward waveform-based measurements also creates significant opportunities for artificial intelligence (AI) and machine learning (ML) integration. High-resolution waveform data is well-suited to ML algorithms, which thrive on rich, minimally processed inputs. AI-driven pattern recognition models, such as convolutional neural networks (CNNs) or deep recurrent neural networks (RNNs), can leverage waveform data to perform anomaly detection, event classification, and predictive analytics with unprecedented accuracy Zhang M. and Liu X. (2021). For instance, AI models trained on waveform data can reliably identify early indicators of equipment failure, distinguish between different fault types, or detect cyber-physical anomalies in real-time Wang and Chen (2020).

Despite these significant advantages, the shift to PoW measurement presents technical and practical challenges. High-resolution sampling requires substantial data storage, communication bandwidth, and computational capabilities Li and Zhou (2021). Transmitting raw waveform data continuously from multiple devices could quickly overwhelm communication networks and central servers. To mitigate these challenges, utilities and technology providers increasingly pursue intelligent data-reduction techniques, such as event-triggered streaming, adaptive sampling, and edge-based waveform analytics, which allow essential waveform features to be extracted locally and selectively transmitted to central analytics platforms Yang and Li (2021).

Further, standardization of waveform-based data collection and exchange formats is still evolving. Efforts such as the IEEE P2650 Standard for Synchro-waveform Data Exchange aim to address these issues by providing guidelines for interoperable waveform data representation, ensuring vendor neutrality, and facilitating seamless integration with advanced analytics and decision-making tools Association (2021).

In summary, the trend toward PoW measurement represents a profound evolution in PMU technology, driven by critical operational needs in modern, dynamic power systems. By enabling accurate characterization of transient phenomena, comprehensive power-quality analysis, and seamless integration with AI-driven analytics, PoW measurements promise to fundamentally enhance grid observability, reliability, and resilience Singh and Sharma (2022).

3 PMU applications

3.1 Wide-area monitoring systems

Wide-Area Monitoring Systems (WAMS) leverage synchronized measurements from geographically distributed PMUs to provide system-level situational awareness across large-scale power networks. By enabling time-aligned observation of voltage, frequency, and power flows, PMU-based WAMS support a wide range of operational functions, including event detection, disturbance analysis, fault location, and topology verification.

In event detection and classification, the high-precision time stamps provided by PMUs allow operators to accurately sequence disturbances such as faults, switching operations, and frequency excursions. Transient phenomena, including lightning strikes, switching surges, and sudden load changes, can be captured and analyzed using synchronized measurements, from which characteristic signatures are extracted to classify events and trigger system alerts Li and Wang (2019). This capability significantly enhances situational awareness compared to conventional SCADA-based monitoring.

For disturbance and fault analysis, high-resolution measurements improve the ability to identify event initiation times, propagation paths, and system responses. PMU-based fault location methods, including impedance-based and traveling-wave algorithms, have been shown to effectively localize faults in complex transmission networks, often achieving spatial accuracy on the order of hundreds of meters Singh and Patel (2016). When combined with point-on-wave (PoW) measurements, traveling-wave-based fault location can achieve even higher accuracy by directly exploiting waveform-level information Zhou and Zhang (2019).

In oscillation monitoring, PMUs enable real-time tracking of electromechanical oscillations, such as local and inter-area modes, which are critical for maintaining grid stability. Complementary measurement infrastructures, including Frequency Disturbance Recorders (FDRs), further enhance oscillation detection and source location capabilities when integrated with PMU data Gupta and Rao (2018). As inverter-based resources introduce higher-frequency and non-stationary oscillatory behaviors, enhanced temporal resolution becomes increasingly important for accurate oscillation detection and characterization.

PMU measurements also support topology verification by identifying inconsistencies between measured electrical quantities and assumed network models. Sudden changes in phase angles or voltage magnitudes can indicate switching actions, breaker misoperations, or unreported outages, allowing operators to cross-validate SCADA-based topology information and improve model accuracy Huang and Zhang (2020).

3.2 PMU-enabled control

Beyond monitoring, PMU data increasingly underpins real-time closed-loop control strategies. In the domain of frequency regulation and fast load control, synchrophasor feedback enables data-driven frameworks that significantly improve response speed compared to SCADA-based methods, as demonstrated by Wu and Palensky (2020). Furthermore, PMUs facilitate Wide-Area Damping Control (WADC) to suppress inter-area oscillations. For example, Dong et al. (2023) utilized PMU measurements to coordinate VSC-HVDC damping, a concept validated by the award-winning GridDamper technology. In the context of renewable integration, PMUs assist grid-forming inverters in phase-locking and synchronizing with weak grids, allowing for dynamic adjustment of voltage and frequency references through AI-enhanced measurement streams Xia and Song (2022).

The integration of PoW-capable measurements further enhances PMU-enabled control by extending system observability to higher-frequency dynamics and faster transients. Improved temporal resolution enables more accurate mode identification, faster disturbance detection, and reduced filtering-induced delays in feedback signals. Consequently, PoW-based measurements can support more responsive and robust wide-area damping and frequency control, particularly in inverter-dominated systems. However, the practical benefits of PoW-enabled control depend on communication latency, edge processing capability, and event-triggered data streaming strategies, which remain active areas of research.

3.3 Illustrative case study: high-resolution event analysis

To illustrate the functional advantages of PoW measurements in capturing fast dynamics, we include an illustrative field case using a representative PoW-capable device (UNDAM UMETER-500). The data were recorded during a grid oscillation event in Seville, Spain, on 11 December 2025.

Figure 2 compares a long-window multi-cycle processing baseline (e.g., 50-cycle moving average) with a 1-cycle waveform-based processing window. The long-window baseline significantly attenuates the oscillatory component and suggests a relatively steady operating condition. By comparison, the 1-cycle processing reveals a distinct 5.88 Hz oscillatory component superimposed on the fundamental trend. This example highlights how waveform-level sensing and short-window processing can improve observability of oscillatory behaviors that may be obscured under coarse temporal resolution, which is increasingly relevant for inverter-dominated grids.

4 Future applications in the AI era

4.1 Challenges from data centers and emerging facilities

The emergence of data centers, high-speed electric vehicle (EV) charging hubs, industrial campuses, and other large-scale, power-electronic-rich facilities presents significant challenges to modern power system operations. These facilities are characterized by high power demand, rapid load variability, and nonlinear electrical behaviors, which introduce new dynamics into both transmission and distribution networks. Traditional measurement infrastructures such as SCADA systems, and even conventional phasor-based PMUs, often lack the temporal resolution and sensitivity required to monitor and manage these rapidly changing conditions effectively.

One of the foremost challenges is the steep increase in both the magnitude and the volatility of electrical demand. Hyperscale data centers, for example, can consume tens to hundreds of megawatts per site, operating continuously with minimal load diversity Smith J. and Lee M. (2020). Their internal operations, including compute workload balancing, thermal management, and periodic switching of backup power systems, can cause rapid fluctuations in active and reactive power consumption. Similarly, large EV charging stations, particularly those employing DC fast chargers, generate substantial and unpredictable demand surges Zhang and Chen (2021). These fast-varying load profiles can induce voltage flicker, frequency excursions, and transformer stress, often beyond what was anticipated in traditional grid planning models.

A second critical issue is the prevalence of harmonic distortion and waveform irregularities introduced by power-electronic interfaces. Facilities such as data centers commonly employ double-conversion Uninterruptible Power Supplies (UPS), high-frequency switching rectifiers, and variable-speed drives, all of which contribute to harmonic pollution and interharmonic resonance Huang X. and Wang L. (2019). These distortions degrade power quality and may propagate upstream, adversely affecting neighboring loads and grid equipment. While traditional PMUs offer some insight into power quality through total harmonic distortion (THD) indices, they are often limited to low-order harmonics and may not fully capture the complex spectral content generated by modern nonlinear loads. In contrast, point-on-wave (PoW) measurement devices provide time-synchronized waveform data with sufficient resolution to enable comprehensive harmonic analysis and real-time detection of waveform abnormalities Patel R. and Kumar A. (2020).

Furthermore, the spatial clustering of such facilities can result in localized stress on substations, feeder circuits, and voltage regulation infrastructure. This phenomenon is particularly evident in suburban and exurban areas where rapid industrial expansion occurs without proportional upgrades to grid assets Kumar and Singh (2022). The fast load dynamics of these facilities may conflict with the slow response characteristics of conventional voltage regulators and capacitor banks, potentially leading to voltage instability or delayed protection operations. Advanced measurement infrastructure, such as PMUs co-located with distribution automation equipment, enables high-resolution monitoring of voltage and current behaviors, thereby supporting dynamic hosting capacity assessments and proactive congestion management Garcia and Kim (2023).

Another important dimension relates to system modeling and load forecasting. Traditional load forecasting techniques rely on historical profiles and statistical models that assume relatively stable load behavior. However, data center operations are increasingly influenced by external factors such as cloud computing demand, artificial intelligence (AI) workloads, and temperature-sensitive cooling requirements, all of which can alter load profiles in real time Lopez and Martinez (2021). PMU data can support the development of physics-informed machine learning models that reflect the true dynamic behavior of such facilities, thereby enhancing forecasting accuracy and operational readiness.

Moreover, the critical infrastructure nature of these facilities raises concerns about grid resilience and cyber-physical security. Abnormal load patterns resulting from cyber intrusions, uncoordinated dispatch of behind-the-meter resources, or improper switching operations within facility microgrids may not be readily detectable using conventional monitoring techniques. PMUs with high-speed waveform streaming capabilities enable early detection of unusual signatures, such as frequency notches, phase angle swings, or sudden harmonic content changes, which may indicate the onset of malicious or unsafe operational behavior Chen and Zhao (2022).

In summary, the integration of data centers and emerging high-density loads into modern power grids introduces multifaceted operational and monitoring challenges. These challenges manifest in the form of fast-changing, non-sinusoidal load behaviors; localized infrastructure stress; degraded power quality; and increased modeling uncertainty. Addressing these challenges requires the deployment of advanced PMU technologies, particularly those capable of point-on-wave measurements, combined with AI-enabled analytics to ensure real-time situational awareness and predictive control. As such facilities continue to proliferate, the strategic placement of PMUs and the intelligent use of their data will be central to maintaining grid stability, resilience, and security.

4.2 AI-enabled applications

The increasing availability of high-resolution PMU and point-on-wave (PoW) measurements has fundamentally changed the scale and structure of data available for power system monitoring and analysis. Traditional model-based techniques, while effective for steady-state and low-dimensional problems, often struggle to fully exploit the high-frequency, high-volume, and non-stationary characteristics of waveform-level data. In this context, artificial intelligence (AI)-enabled analytics are becoming an essential complement to conventional approaches.

In this review, AI is used as an umbrella term for data-driven inference and decision-making frameworks, whereas machine learning (ML) refers to specific learning models within this broader category. Recent studies have demonstrated that deep learning architectures, particularly Convolutional Neural Networks (CNNs) and autoencoders, are effective in extracting informative features from high-dimensional waveform data streams Mestav et al. (2023); Wang and Li (2020). In addition, attention-based models such as Transformers are being actively explored for capturing long-term temporal dependencies in time-series forecasting. Although their application to sub-cycle power system dynamics is still evolving, early results suggest strong potential for analyzing complex, non-stationary behaviors Zeng et al. (2023).

A prominent application of AI-enabled PMU analytics is real-time inertia and stability assessment in low-inertia power systems. For example, Jiang and Lee (2024) presented an AI-enhanced framework for estimating system inertia using high-resolution PMU measurements, enabling more accurate and timely assessment of stability margins under renewable-dominated operating conditions. Similar data-driven approaches have been reported for online oscillation detection and mode identification, where CNN-based and attention-based models leverage temporal and spectral patterns embedded in synchronized PMU and PoW measurements Singh and Agarwal (2024). Furthermore, generative AI architectures are emerging as a promising solution to the dual challenges of data volume and protection reliability in stochastic grids. An innovation autoencoder framework proposed in Tong et al. (2024) learns statistical representations of continuous PoW waveforms, enabling efficient data compression to mitigate bandwidth constraints while simultaneously supporting robust sequential fault detection. This approach is particularly effective in addressing complex protection issues such as sympathetic tripping and protection blinding in inverter-dominated distribution networks.

Beyond purely data-driven learning, physics-informed learning frameworks represent an important direction for enhancing the reliability and interpretability of AI-based power system analytics. Physics-Informed Neural Networks (PINNs) explicitly incorporate physical laws, including swing equations and network constraints, into the learning process, thereby improving generalization under limited or noisy measurement conditions Raissi et al. (2019). Recent studies have explored the application of PINNs to dynamic parameter estimation and stability assessment using PMU data, demonstrating their potential to bridge first-principles modeling and data-driven inference Zhao and Liu (2022).

From a system-level perspective, the integration of AI-enabled analytics with PoW-capable PMUs has important implications for smart grid observability and control. Modern smart grids, characterized by high penetration of distributed energy resources, microgrids, and power-electronic interfaces, impose increasingly stringent requirements on measurement resolution and situational awareness. Waveform-level measurements provide enhanced visibility into fast dynamics and localized phenomena that are difficult to capture using conventional phasor data alone. When combined with edge computing architectures and intelligent data reduction strategies, AI-assisted PoW analytics enable scalable, real-time monitoring and support advanced functions such as anomaly detection, adaptive protection, and data-driven situational awareness. These capabilities are expected to play a critical role in future wide-area monitoring and control systems, as well as in the broader development of smart cities and AI-driven energy management platforms.

5 Conclusion

Phasor Measurement Units (PMUs) continue to serve as a cornerstone technology in the evolution of modern power systems and smart grids. Advances in PMU hardware, signal processing algorithms, and system architectures have significantly expanded their role from steady-state monitoring toward capturing increasingly complex grid dynamics. In particular, the transition from conventional phasor-based measurements to point-on-wave (PoW) sensing enables visibility into fast electromagnetic transients, high-frequency oscillations, and inverter-driven interactions that are not adequately represented by traditional phasor abstractions. This shift marks a fundamental enhancement in system observability and lays the foundation for more informed monitoring and control strategies in low-inertia, inverter-dominated grids.

Looking forward, the integration of high-fidelity PMU and PoW measurements with artificial intelligence techniques presents substantial opportunities to improve grid reliability, resilience, and operational intelligence. As power systems become more complex and cyber-physical interactions intensify, AI-enabled analytics are increasingly necessary to manage the volume, velocity, and variety of waveform-level data. To fully realize this potential, several key enablers must be addressed, including interoperable waveform data standards, deployment-oriented performance metrics that explicitly account for latency, reliability, and cybersecurity, and effective knowledge transfer from academic research prototypes to operational wide-area monitoring systems (WAMS). Collectively, these efforts will ensure that PMUs evolve beyond measurement devices and become integral components of data-driven power system operation and innovation.

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: Terence O’Donnell, University College Dublin, Ireland

Reviewed by: Stelios Ioannou, University of Central Lancashire, Cyprus

Chisom Ogbogu, Carnegie Mellon University, United States