Deep Q-Network Reinforcement Learning-Based Rotor Side Control System of a Grid Integrated DFIG Wind Energy System Under Variable Wind Speed Conditions

Wind power generation is a sustainable way to meet rising energy demand. Fluctuating wind speed causes fluctuating output power, which threatens power system stability. Due to the wind energy conversion system's (WECS) output power transients, the conventional control system has ineffective invariance against power system reservations. Overshoot, settling time, gain limitations, and steady-state error degrade power system stability and must be minimized to zero. This paper introduces modern controller design and implementation. The research used five controllers, FOPI, FUZZY, HYBRID FOPI & FUZZY, Adaptive Neuro-Fuzzy Inference System (ANFIS), and Deep Q-Network (DQN) to optimize WECS efficiency by regulating DFIG rotor current. First, a FOPI controller is created and implemented to minimize steady-state errors and enhance system efficiency. To reduce overshoot, a fuzzy controller is created and implemented. An adequate system output controller is achieved by combining FOPI and Fuzzy methodologies, followed by ANFIS. A hybrid adaptive DQN controller is developed and applied to regulate the system under constant and variable wind speeds. The effectiveness of the hybrid adaptive DQN controller is evaluated based on its ability to reduce transient harmonic distortions (THDs), percentage overshoot, settling time, and steady-state inaccuracy in rotor and stator current transients. Compared to the ANN, hybrid fuzzy-FOPI controller, fuzzy controller, and regular FOPI controller, the Hybrid adaptive DQN controller has more excellent transient responsiveness, torque control, and maximum power extraction efficiency due to its ability to overfit due to the added layers of abstraction, which allow it to model rare dependencies in the training data.