Resiliency of Power Systems in the Cyber Layer: Application to Load Forecasting and Security Assessment

Integrating sensors, actuators, and communication infrastructure in the electrical grid creates a smart grid, known as a cyber-physical system (CPS), which combines the physical framework with a cyber layer. The cyber layer is crucial as it houses the decision-making responsible for reliable operation. However, the complexity of the physical layers, due partly to the deployment of integrated battery resources (IBRs), and the cyber layer itself, introduces challenges such as reliance on measurement quality and vulnerability to data corruption from cyber threats. These challenges result in uncertainties in the CPS framework, emphasizing the need for accurate and robust responses from the power grid's cyber layer through cyber resiliency. To establish resilient responses, the concept of resiliency and its framework are defined in the context of CPS. Among the functions ensuring the performance and reliability of the power grid, two specific tasks, namely short-term load forecasting (STLF) and security assessment (SA), are examined.

STLF, used for scheduling and assessing power system security, faces challenges with misleading forecasting, particularly anomalies caused by temperature data. To address this, STLF is explored. Similarity measures and a distributed mining of lag correlation approach are utilized to analyze load behavior and detect anomalies. This approach incorporates physical characteristics, making it difficult for malicious actions. The effectiveness of this method is validated using historical load data from the Electric Reliability Council of Texas (ERCOT).

SA evaluates the risk of disruptions in the system operating continuously. Given operational uncertainties and measurement challenges, developing an efficient SA model capable of capturing most scenarios is crucial. Representing the SA task in a graph format enables leveraging connectivity and topological information obtained through centrality measures. A resilient model based on graph neural networks (GNN) is constructed to aggregate data across the grid. The framework's efficacy is demonstrated on the IEEE 118-bus system for voltage stability SA (VSSA) task under various operating points and stress conditions. Comparative analysis, including the evaluation model capacity, highlights the superiority of the proposed method over a neural network approach.