Short-term load forecasting for integrated energy systems based on deep learning networks and evolutionary algorithm

This study proposes short-term load forecasting scheme for integrated energy systems based on deep learning networks and evolutionary algorithm, in which the CNN-BiGRU-Attention model that integrates variational mode decomposition and crested porcupine optimization is presented for load forecasting. Extracting spatial features of load data through CNN (Convolutional Neural Networks, CNN), capturing temporal features through BiGRU (Bidirectional Gated Recurrent Unit, BiGRU) network, and enhancing key information weights through attention mechanism. Using variational mode decomposition technology to perform modal decomposition on the original load effectively eliminates random noise and extracts trend terms and harmonic components. Innovatively introducing crested porcupine optimization algorithm to adaptively optimize hyperparameters such as training times, number of neurons, and learning rate of BiGRU. The simulation results show that compared with the current prediction model, the proposed model has better load forecasting accuracy and stable error distribution, indicating that the model has superior performance. Verified the effectiveness of the model in spatiotemporal feature fusion and parameter optimization.