Short-Term Charging Load Forecasting for Electric Vehicles Based on User Portraits and LSTM Model

With the growing global call for environmental protection and sustainable development, electric vehicle (EV), as an important part of new energy transportation, are rapidly gaining popularity. However, the time volatility of EV charging load poses significant challenges to the power grid. To address the uncertainty and volatility of EV charging load, this paper proposes a hybrid forecasting model based on K-means clustering and Long Short-Term Memory (LSTM) networks. First, the K-means clustering algorithm is used to classify EV charging behavior and identify different user groups' charging patterns. Then, the LSTM model is applied to each class of data for time series forecasting, capturing the long-term dependencies and nonlinear features of the charging load. To further optimize the model's performance, the Particle Swarm Optimization (PSO) is introduced to automatically adjust the hyperparameters of the LSTM model. Experimental validation using electric vehicle charging data from the California Institute of Technology demonstrates that this method effectively improves the accuracy and adaptability of charging load forecasting, providing strong support for load management in smart grids.