Electric Vehicle Charging Load Forecasting Based on Sequence Optimization Decomposition and Reconstructed Bidirectional Long Short-Term Memory Neural Networks

In response to the impact of electric vehicle charging loads on the power grid, this study proposes a method based on sequence optimization decomposition and a bidirectional long short-term memory neural network for enhanced forecasting accuracy. The method involves first combining ensemble empirical mode decomposition with the differential evolution optimization algorithm to determine the optimal decomposition parameters, thereby reducing information loss during sequence decomposition. The load time series is then decomposed into multiple intrinsic mode functions and a residual. Sample entropy is calculated for each component, and based on these values, K-means clustering is employed to obtain the optimal number of clusters for the reorganized components. The reorganized components are then input into the neural network prediction model to obtain individual prediction results, which are linearly combined to produce the final load forecast. The model is trained using historical data from a charging station located in Hefei, and it conducts load forecasting post-training. A comparative analysis demonstrates that the proposed method achieves improved accuracy in forecasting electric vehicle charging loads.