Optimized Controller Design for Renewable Energy Systems by using Deep Reinforcement Learning Technique

Abstract

Power networks are currently evolving in a more inventive way as a result of the absorption of renewable energy sources. As a result, it is now considerably harder for energy networks to control the frequent variations in connector voltage and frequency. Deep networks and probabilistic reasoning are used in this study to improve scenario planning, capacities, and microgrid regulation at the moment (solar and batteries). The goal of this rule is to reduce costs while increasing overall effectiveness. Every time shifting renewables, shifting demands, and autos are combined with inverters, the complexity of the problem increases. As a result, the solution is more sincere. The proposed method modifies the needs of the proportional-integral-derivative (PID) controllers using an upgraded triple delay deep course gradients (TD3) based hybrid learning agent. Each representation is given a non-negative linked layer that is also provided with objective functions in order to prevent the occurrence of improper gain. Each agent uses the knowledge of the local area control error to minimize changes in frequency and tie-line power that arise between them. The deep learning systems are taught to obtain the ideal step size for the specified two-area linked system. The controller gains are calculated using the integral error value of the controllers' error as a weighting factor, but these gains are later updated and optimized using fuzzy logic. The controller error is where this process starts. The given approach is put through scenarios with irregular load generation disturbances and nonlinear generating features in order to assess its effectiveness. The simulation's results show that the suggested strategy is superior to previous ones that were described in the research and show that it can effectively handle nonlinearities brought on by variations in load generation. This was shown by the fact that it handled nonlinearities brought on by variations in load generation successfully. Deep learning and machine learning techniques are used to calculate the power variations ratio in accordance with the dynamic load fluctuations in order to accomplish this goal.