Search

Artificial intelligence is increasingly used for real-time decision support and control in cyber-physical smart grids. However, the growing reliance on data-driven control creates a need for methods that are both explainable to human operators and consistent with electrical-network physics. This review proposes a two-dimensional classification scheme for AI methods in smart-grid control. The scheme combines an explainability dimension with a physics-informed dimension and applies the resulting framework to voltage regulation, congestion management, and microgrid control. A systematic literature search identified 25 peer-reviewed journal articles published between 2017 and 2025. The selected studies were classified according to their explainability mechanism, degree of physical-knowledge integration, application domain, and relevance to cyber-physical grid operation. The review shows that most studies emphasize either explainability or physical consistency, while methods combining both remain limited. Voltage regulation is dominated by reinforcement-learning and safety-constrained control studies, whereas microgrid control contains a broader variety of AI methods. The proposed classification provides a structured basis for analysing trustworthy AI in smart-grid control. It highlights methodological gaps and identifies under-developed combinations of explainability, physics-informed learning, and operational deployment.

Autonomous control is increasingly relevant to smart-grid operation, where control systems may support voltage regulation, frequency control, dispatch, and anomaly response. Yet their trustworthiness remains uncertain when decisions are opaque, intervention options are unclear, and cyber-physical conditions are unstable. Existing research often treats explainability, operator intervention, and cyber-physical uncertainty as separate design or evaluation issues. This separation limits theory development because trust in autonomous grid control depends on how these mechanisms operate together. This article develops a conceptual theory of trustworthy autonomous smart-grid control. The theory links explainability, operator intervention, and cyber-physical uncertainty to both perceived and operational trustworthiness. The article uses theory-building logic rather than empirical testing. It synthesizes concepts from trustworthy AI, human-automation interaction, explainable AI, and cyber-physical security to construct mechanisms and derive propositions. The proposed model specifies causal pathways through which explanation quality, intervention capability, and uncertainty handling shape trustworthiness. It also identifies interaction effects, boundary conditions, and testable propositions for future research. The theory provides a foundation for designing, evaluating, and empirically testing trustworthy autonomous control systems in smart-grid environments. It emphasizes that trustworthiness is not produced by transparency, human oversight, or robustness alone, but by their coordinated operation.

Foundation models are transforming artificial intelligence by enabling pre-training, adaptation, and reuse across heterogeneous tasks. Their application in energy systems is emerging rapidly, creating a need for systematic synthesis across forecasting, control, and decision-support settings. This systematic review identifies, categorizes, and evaluates the use of foundation models in energy-system forecasting, grid-edge control, and smart-grid decision support. It focuses on how large language models, pre-trained transformers, multi-modal models, and emerging time-series foundation models are adapted for energy tasks. A PRISMA-guided search of peer-reviewed literature from 2017 to 2025 resulted in 30 included studies. Models were classified by architecture, fine-tuning method, data modality, and application domain. The review finds that large language models dominate decision-support applications, while pre-trained transformers and time-series foundation models are most developed in forecasting. Grid-edge control applications remain nascent, with most studies addressing planning, diagnostics, or dispatch rather than closed-loop physical deployment. Foundation models hold significant promise for smart grids, but critical gaps remain in real-time control, safety assurance, privacy, and evaluation. A research agenda is proposed to guide rigorous, trustworthy, and operationally relevant deployment.

Smart-grid optimization increasingly uses artificial intelligence, automated control, and data-driven market coordination to manage low-carbon electricity systems. These developments create new opportunities for flexibility, but they also raise unresolved questions about equity, accountability, and justice. Existing reviews often examine energy justice, smart-grid optimization, algorithmic governance, and flexibility markets as separate domains. This fragmentation limits understanding of how optimization models and market rules may reproduce or reduce social inequalities. This article critically reviews the literature on justice-aware smart-grid optimization and develops a structured future research agenda. It focuses on how energy equity, algorithmic governance, and low-carbon flexibility markets can be integrated into future smart-grid research and policy. A targeted review of 26 peer-reviewed articles published between 2017 and 2025 integrates perspectives from energy justice, smart-grid studies, AI governance, and electricity market design. The review identifies conceptual tensions, methodological gaps, and governance challenges rather than presenting new empirical data. The review shows that justice is often operationalized narrowly in optimization, while broader concerns about recognition, participation, and accountability remain underdeveloped. It also finds that flexibility markets can distribute benefits unevenly unless market design explicitly addresses barriers to access and agency. A more interdisciplinary and justice-centered approach is needed to guide smart-grid optimization. The proposed agenda calls for stronger integration between technical modeling, energy justice theory, participatory governance, and regulatory innovation.

Autonomous smart grids increasingly rely on AI-driven decisions, yet legal frameworks have not kept pace. Liability for autonomous actions remains ambiguous, threatening both innovation and accountability. No existing conceptual framework systematically links the algorithmic decision layer, operator oversight, and regulatory auditability to address the liability gap in autonomous grid operations. This creates uncertainty when harmful outcomes arise from distributed interactions among software, operators, vendors, and regulated utilities. This article develops a conceptual framework that connects these three layers into a coherent liability-aware governance architecture for autonomous smart grids. The aim is to clarify how responsibility could be assigned, reviewed, and operationalized before and after autonomous grid actions occur. The framework is built through conceptual synthesis of AI law, human-in-the-loop research, and regulatory governance literature. No experiments are conducted, and no quantitative data are presented. The framework defines the roles and interactions of algorithmic decisions, operator oversight, and auditability mechanisms, and derives eight testable propositions. Two tables summarize the decision-liability mapping and the framework logic. The liability-aware framework provides a foundation for designing accountable autonomous grid systems and for future interdisciplinary legal-technical research. It supports legal clarity while preserving the operational benefits of intelligent grid automation.

Resilience assessment in smart grids is often limited by the small number of hand-crafted scenarios. Generative AI can create diverse, realistic synthetic stress events, providing a more rigorous benchmark. This article presents a generative scenario benchmark that combines synthetic cyber-attacks and extreme weather events with a standardized set of DER recovery actions to evaluate grid resilience. The benchmark is designed for comparative testing of cyber-physical distribution-grid controllers. The benchmark uses generative models to produce a large library of attack and weather scenarios. DER recovery actions are catalogued, and several baseline controllers are evaluated using standardized resilience metrics on a distribution feeder. The benchmark reveals distinct performance profiles: reinforcement learning excels in speed, robust optimization improves worst-case protection, and rule-based control provides simple but conservative responses. Three tables summarize the attack types, recovery actions, and comparative performance. The proposed benchmark offers a scalable, reproducible, and diverse testbed for developing and comparing next-generation resilient grid controllers. It supports transparent evaluation under coupled cyber-physical stress rather than isolated contingency cases.

Over the past few years, integrating Distributed Energy Resources (DERs) with advanced communication infrastructures has created major challenges for the control and protection of electrical power systems. These difficulties stem mainly from the rapid evolution of smart grids and the growing threat of cyberattacks. As grid-tied solar Photovoltaic (PV) installations—equipped with smart inverters—become more widespread, their exposure to malicious cyber activities and the resulting risks to power network stability have become a pressing issue, especially given the inverter control logic. The present study examines the outcomes of targeted cyber intrusions targeting key PV system elements, with particular attention to the inverter and the Overcurrent Protection Relay (OCR). It further analyzes how these two components interact across a range of cyber-attack conditions. Simulations are conducted using a three-phase radial Electromagnetic Transients Program (EMTP) model to represent the grid and study transient responses during various cyber intrusions. Analysis results uncover the tight coupling between weaknesses in inverters and protection relays, demonstrating how disruption of one can severely affect the performance of the other. The work also quantifies how cyber incidents degrade the efficiency and operational reliability of PV systems connected to the grid. As a concrete illustration, when PV inverters were subjected to a brief 0.1 s pulse signal attack, the OCR failed to detect the abnormality and isolate the fault. This breakdown caused pronounced harmonic distortion together with considerable active power losses, stemming directly from the protection scheme’s inability to recognize and counteract the irregular disturbance. Overall, this investigation provides important new insights into cybersecurity challenges in grid-integrated solar PV installations. It underscores the urgent need to adopt stronger defensive strategies and build greater system resilience as power networks continue evolving into intelligent, interconnected smart grids.

The Smart Grid is an advanced electricity network deeply intertwined with cyber technologies that oversees power generation, long-distance transmission, and final delivery to end users. At its core lies the substation. However, embedding digital systems within substations has dramatically widened the range of potential vulnerabilities. Advanced adversaries, such as those behind the PipeDream APT, deliberately exploit several industrial protocols, including Modbus, DNP3, and IEC61850. Substations face strict operational constraints that complicate the adoption of conventional protective mechanisms, such as data encryption or network intrusion detection systems. To address these difficulties, we present an integrated, trust-centric architecture that significantly strengthens substation defenses. The proposed architecture integrates three main components: a trust evaluation model, a dynamic risk assessment model, and a trust portability model. The trust evaluation model can uncover protocol-specific malicious activities targeting Intelligent Electronic Devices and SCADA human-machine interfaces. The dynamic risk assessment model continuously determines the substation’s current security risk exposure. The trust portability model analyzes whether a device, together with its accumulated trust profile, can be effectively migrated and reused in another substation environment. For a realistic evaluation, we constructed a containerized (Docker) substation replica organized around a multi-agent framework and equipped with a live dashboard modeled after modern Security Operations Centers. Effectiveness was measured by simulating a broad spectrum of adversary behaviors aligned with the MITRE ICS ATT&CK matrix. Experimental results indicate robust resistance to various attack vectors, reliable detection of malicious actions, and appropriate incentivization of well-behaved, trustworthy components. 

With the rapid expansion of urban power grids and the widespread adoption of renewable energy sources, conventional fault-diagnosis methods for power grids face significant challenges in handling the complexities of modern, highly interconnected systems. While artificial intelligence approaches offer promising alternatives for power grid fault diagnosis, the scarcity of labeled grid data severely hinders the application of AI techniques in this domain. To tackle these issues, this study introduces a semi-supervised learning framework featuring self-supervised learning and an adaptive threshold mechanism (SAT-SSL) designed for fault detection and classification in power grids. Compared with existing approaches, the proposed method substantially reduces reliance on labeled data while maintaining excellent recognition performance. First, a frequency-domain analysis is applied to power grid data to identify and filter anomalous events, which are then classified and labeled based on their visual characteristics, thereby constructing a dedicated power grid dataset. Next, the Yule–Walker algorithm is employed to extract features from the power grid data. A semi-supervised learning framework is subsequently developed, integrating a self-supervised loss and a dynamic threshold mechanism to enhance the model’s information extraction capacity and adaptability across diverse operational scenarios. Finally, the constructed power grid dataset, together with two standard benchmark datasets, is utilized to evaluate the model’s effectiveness. Experimental results demonstrate that the proposed model maintains a low error rate across multiple scenarios and varying label proportions. Specifically, on the power grid dataset, retaining only 5% of the labels yields an error rate of just 6.15%, confirming the method’s ability to achieve precise fault detection and classification even with very limited labeled data.

The shift of electrical power systems toward intelligent smart-grid architectures delivers better dependability, operational performance, and real-time visibility, yet simultaneously heightens exposure to cyber threats. Thanks to the massive amounts of data now generated in smart grids, machine learning techniques have emerged as a particularly promising avenue for bolstering defensive capabilities. While the strengths of machine-learning solutions for cybersecurity improvement are clear, these same models form part of the attack surface and remain especially susceptible to carefully crafted adversarial perturbations. This work analyzes the resilience of autoencoder-based intrusion detection mechanisms in smart grids against adversarial examples. We begin by presenting an original iterative procedure for synthesizing adversarial instances. Next, we demonstrate that a white-box adversary with full knowledge of the autoencoder detector can reliably produce stealthy, misclassifying samples using the proposed approach. Experimental results show that simply starting from basic adversarial seeds is insufficient to generate effective evasions, underscoring the considerable difficulty of constructing potent adversarial examples targeting autoencoder-based anomaly detectors deployed in smart-grid environments.

A wide range of strategies has been proposed to identify and prevent non-technical electricity losses resulting from fraudulent activities. Among these, machine learning algorithms and data-driven methods have become increasingly prominent compared to conventional approaches, owing to their superior performance, leading to growing adoption in recent years. This paper introduces a novel two-step method for detecting fraudulent Non-technical losses (NTLs) in smart grids. The initial step is to transform the time-series data by incorporating additional features extracted from the publicly available State Grid Corporation of China (SGCC) dataset. These features are derived following the detection of sudden changes in electricity consumption patterns, using the sum of finite differences, the Auto-Regressive Integrated Moving Average model, and the Holt-Winters model. Subsequently, five classification models are applied to train and evaluate a fraud detection system using the SGCC dataset. The performance evaluation shows that the Gradient Boosting Machine delivers the best results among the five models. This two-step procedure enables classification models to outperform previously reported top-performing techniques in terms of accuracy, F1-score, and other key metrics for non-technical loss detection.

Modern smart grids aim to deliver exceptional reliability and operational flexibility in power delivery. Contemporary electrical networks rely on synchronized wide-area monitoring tools, notably Phasor Measurement Units (PMUs), to supply instantaneous system status updates. These devices support prompt problem identification and corrective actions while ensuring uninterrupted grid operation. The surge in available measurement data has accelerated the adoption of machine learning techniques for tasks such as power system protection and state estimation. This work presents an intelligent fault detection and classification (FD&C) framework for transmission lines that employs an auto-encoder neural network architecture. The study evaluates and contrasts the capabilities of a Multi-Layer Extreme Learning Machine (ML-ELM) with those of a Stacked Auto-Encoder (SAE) neural network. These models are further benchmarked against established classifier approaches that rely on wavelet-based feature extraction, as well as against various other deep learning strategies. The proposed auto-encoder solutions attain perfect (100%) fault detection rates and achieve fault classification accuracies of 99.92% and 99.79%, all while demanding far less computation time during the testing phase. The ML-ELM variant stands out for its efficiency, maintaining excellent classification performance with both training and testing durations below 50 milliseconds. To reflect practical field conditions, the models were rigorously tested on datasets incorporating additive noise levels corresponding to signal-to-noise ratios (SNR) between 10 dB and 40 dB. Validation was performed using simulated data from the IEEE 39-bus benchmark system.

The Smart Grid (SG) seeks to overcome the limitations of conventional power systems by integrating renewable energy sources. It leverages wireless sensor networks (WSNs) to improve monitoring of power generation equipment. However, the open and expansive deployment of WSNs in SG introduces new vulnerabilities and potential attacks. This paper introduces a novel security framework for WSNs in SG that utilizes public key infrastructure (PKI). The core concept involves distributing the duties of the certification authority (CA) across multiple sensor nodes to guarantee service availability and scalability. A new lightweight clustering method is developed to select these nodes, accounting for their trust values and remaining energy. This framework delivers essential security features, including authentication and data confidentiality, while countering threats such as Sybil attacks and eavesdropping. Comprehensive simulations with OMNET++ and the Castalia framework demonstrate the clustering algorithm’s effectiveness: nearly all sensors join clusters, and the number of cluster heads remains consistent even when malicious nodes are present, highlighting the framework’s resilience.