Determination Method of Optimal Reserve Margin Based on Explainable AI Using Gaussian Process Regression Model and SHAP

Abstract

Electric power systems with increasing photovoltaic (PV) systems face concerns regarding degradation in frequency stability due to heightened output forecast errors. As a countermeasure, given the dynamic factors like demand, PV output, and meteorological elements, calculating the optimal reserve margin (ORM) becomes crucial for economic efficiency and resilience reinforcement. To ensure an efficient ORM, Artificial Intelligence (AI) is one of the useful strategies used to analyze the combination of all the elements. However, AI is characterized by a black box problem, and to achieve transparency, AI needs to be transformed into explainable AI. To begin with, this paper analyzed all features importance using SHapley Additive exPlanations (SHAP), adopting a Gaussian process regression model. Then, relevant explanatory variables were selected to improve the prediction accuracy of the ORM. Finally, to verify the effectiveness, this paper planned day-ahead scheduling while securing the ORM determined by the proposed method. It executed detailed demand/supply and system frequency simulations as an operation. The proposed method decreased the risk posed by PV output forecast errors and shortage of reserve margin. Also, the maximum PV capacity increased from 96.2% to 166.2% while maintaining frequency stability.