Reactive power control in photovoltaic systems through (explainable) artificial intelligence

Across the world, efforts to support energy transition and halt climate change have resulted in significant growth of number renewable distributed generators (DGs) installed over last decade, among which photovoltaic (PV) systems are fastest growing technology. However, high PV penetration electricity grid is known lead numerous operational problems such as voltage fluctuations line congestions, could be eased by utilizing reactive power capability systems. To this end, we propose use artificial neural network (ANN) predict optimal dispatch learning approximate input–output mappings from AC flow (ACOPF) solutions either a centralized or decentralized manner. In case control, leverage Shapley Additive Explanations (SHAP), an explainable intelligence (XAI) technique, identify non-local state measurements significantly influence each individual system. Both ANN-based controllers evaluated through study based on CIGRE medium-voltage distribution compared baseline control strategies. Results show that both exhibit superior performance, hindering congestions encountered with strategies while recording saving 0.44% fixed factor control. By leveraging ANN SHAP, proposed for able achieve ACOPF-level performance promoting data privacy reducing computational burden.