Attention-Guided Policy Gradient Reinforcement Learning for Autonomous Fault-Aware Control in Wind Energy Systems
Abstract
Renewable Energy Sources are a sustainable and reliable form of energy that is utilized for energy generation, which impacts the economic status of the Country. Wind RES depends on the wind energy, where the process of aerodynamic conversion and mechanical to electrical energy conversion takes place. The installation and maintenance costs of RES are very high, but they have great advantages. It is important to diagnose the fault proactively to reduce the occurrence of major failures. Hence, the proposed work uses the SCADA-based fault detection systems using the Luong attention mechanism, and the control mechanism is optimized using the Policy-based Gradient Reinforcement Learning (RL) algorithm. The proposed work was evaluated using the Wind Turbine Dataset of SCADA from the Turkish wind turbine, which showed an accuracy of 97.65% with an F1 score of 97.3, and RL optimization showed a reward percentage of 86% for the actions taken based on the policies based on the fault diagnosis.