Microgrid energy management strategy based on LSTM-DDQN

Abstract This paper introduces a novel algorithm, LSTM-DDQN, designed to address the stochastic economic dispatch problem in microgrids. The inherent variability and randomness of photovoltaic (PV) energy pose challenges for accurate output prediction. Traditional approaches, including physical and statistical methods, often fall short in terms of accuracy. For microgrids with a high proportion of PV generation, we propose a new algorithm that combines Long Short-Term Memory (LSTM) neural networks with the Double Deep Q-Network (DDQN) algorithm. The algorithm leverages the LSTM model to capture the uncertainty of the learning environment and optimizes the Q-value iteration rules in the DDQN algorithm, thereby enhancing the training speed of the neural network. Experimental results demonstrate the algorithm’s remarkable dispatch capabilities in managing the interplay between PV generation, battery capacity, and overall load demand. This algorithm provides robust support for the stable operation of microgrids.