Short-Term Load Forecasting for Wind and Photovoltaic Grid Integration Using Multi-Agent Deep Reinforcement Learning

Due to the widespread integration of wind and photovoltaic power into the grid, its randomness and uncertainty directly affect the stability of the power system, and its grid structure and operation mode have become more complex. Traditional single prediction models are difficult to comprehensively analyze multiple influencing factors. To improve prediction accuracy, a regional short-term load forecasting (STLF) method based on multi-agent deep reinforcement learning is proposed. First, based on the Mahalanobis distance clustering algorithm, the bus load is divided into three types: conventional bus, wind and photovoltaic grid-connected bus, and impact load bus. At the same time, to better describe meteorological conditions, the multivariate RapidMIC algorithm is used to select factors such as climate temperature, light, and wind power that are closely related to the load. Then, the policy network and value network of the multi-agent deep deterministic policy gradient (MADDPG) algorithm are improved, and the sparrow search algorithm is introduced to optimize the hyperparameters to obtain the optimal MADDPG algorithm. Finally, the three types of bus loads and influencing factors are input into the improved MADDPG algorithm for training and learning to obtain STLF results for the region. Based on load data from a certain jurisdiction, the proposed method is experimentally tested, and the results show that its predicted MAE and MAPE values are 7.86[Formula: see text]MW and 1.17%, respectively, with a prediction time of 11.43[Formula: see text]s. Overall, it has better prediction performance.