EV Charging Load Forecasting Based on Improved SOM Clustering and ECA‐LSTM‐SA Model

With the rapid increase in the number of electric vehicles (EVs), the surge in stochastic EV charging loads poses significant challenges to the secure and stable operation of distribution networks. To address this issue, accurate EV charging load forecasting is crucial. This paper proposes a novel EV charging load forecasting method based on an improved self‐organizing map (ISOM) clustering algorithm and an ECA‐LSTM‐SA model. First, an ISOM clustering algorithm is developed, which takes load values at different time instants as inputs, employs cosine similarity as the distance metric, and utilizes particle swarm optimization (PSO) to determine hyperparameters. This clustering method is used to segment charging loads into similar daily patterns. Then, an ECA‐LSTM‐SA forecasting model is constructed by integrating efficient channel attention (ECA) and soft attention (SA) mechanisms with long short‐term memory (LSTM) networks to enhance feature extraction and prediction accuracy. The model is trained and used for forecasting within each clustered dataset. Finally, the prediction results from all clusters are concatenated to obtain the final forecasted charging load. Simulation results demonstrate that the proposed method significantly improves prediction accuracy and stability.