1. Improved reinforcement learning strategy of energy storage units for frequency control of hybrid power systems
- Author
-
Yakout, Ahmed H., Hasanien, Hany M., Turky, Rania A., and Abu-Elanien, Ahmed E.B.
- Abstract
An improved Reinforcement Learning (RL) agent with a Deep Deterministic Policy Gradient (DDPG) algorithm is proposed to control the frequency of hybrid power systems. Weighted signals of system frequency, frequency deviation, integration of frequency deviation, and differentiation of frequency deviation represent the inputs to the RL system. The reward equation used is a discrete equation considering a study period of 400 s with a sampling time of 0.25 s (i.e., 1600 samples per run). The reward equation allows the RL method to consider the solutions where the system operates with a frequency between 49.8 and 50.2 Hz, which is the allowable operating range of frequency for a 50 Hz-based power system under normal load changes in several countries. To enhance the performance of the proposed RL controller further, the input weights (i.e., observation weights) are optimized by the Gorilla Troops Optimizer (GTO) algorithm, which comprises high convergence and robustness features. The proposed control technique is verified using a two-area system with various combinations of renewable energy sources, conventional sources, and energy storage units. The energy storage units include battery energy storage and superconducting magnetic energy storage. This article's main contribution is applying a novel GTO-based optimal RL controller to enhance the frequency control of hybrid power systems. The proposed load frequency controller is compared with optimally tuned PI and PID controllers. The results of this comparison are in favor of the proposed controller. All studies of this research are carried out using MATLAB environment.
- Published
- 2023
- Full Text
- View/download PDF