Ultra Short-Term Charging Load Forecasting Based on Improved Data Decomposition and Hybrid Neural Network

Ultra-short-term charging load prediction is of crucial importance for the real-time scheduling and stable operation of power systems. To address the problem of the strong nonlinearity of power load sequences, this study proposes an ultra-short-term charging load prediction model that combines a Time Convolutional Network-Bidirectional Gated Recurrent Unit (TCN-Bi-GRU)and uses the Osprey Optimization Algorithm (OOA) based on the opposition-based learning strategy to optimize Variational Mode Decomposition (VMD). First, an OOA incorporating an opposition-based learning strategy was introduced to optimize the key parameters of the VMD. The permutation entropy is utilized as the fitness function to iteratively obtain the optimal combination of fitness parameters.Secondly, VMD with the optimal parameter combination is used to decompose the original charging load sequence into multiple relatively stable components, reducing the non-stationarity and complexity of the sequence. Then, the components and external data are divided into training sets to train TCN-Bi-GRU. After the model training was completed, the test sets of each component were predicted separately to fully explore the spatiotemporal characteristics of the dataset. The experimental results show that compared with single models, the proposed model performs better in terms of Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and R-squared in three different scenarios, proving that the model has high prediction accuracy and good robustness in ultra-short-term charging load prediction.