A Novel Deep Reinforcement Learning and Trust Region Policy Optimization based Anomaly Detection in Solar Power Generation

Anomalous conditions in photovoltaic (PV) systems cause major interruption to the system normal operation and are the main reason for power loss. To avoid these consequences, detection and analysis of faults in PV systems must be enhanced effectively and promptly. In this paper, a fault detection and classification model based on deep reinforcement learning (DRL) is developed and improved. The required datasets are collected using a real laboratory PV array. The fault detection and classification environment in DRL is built based on the Markov Decision Process (MDP), and the DRL agent is trained based on the Trust Region Policy Optimization (TRPO) algorithm. The proposed model is then implemented to detect and classify line-to-line (LL) and degradation faults in a PV array and the results confirm an average accuracy of 99.84% under various conditions.