Short Term Load Forecasting of Integrated Energy System Based on CNN-BiGRU-Transformer Model

Accurate load forecasting is crucial for the economic dispatch and efficient operation of integrated energy systems. We propose a short-term forecasting method in this paper for multi-dimensional loads in integrated energy systems based on the CNN-BiGRU-Transformer (convolutional neural networks - bidirectional gated recurrent unit - Transformer) model. Firstly, Pearson correlation coefficient (PCC) is used to select the optimal input features. Secondly, using the variational mode decomposition method, the original historical load is decomposed into multiple intrinsic mode functions to eliminate random noise and extract trends and harmonics. On this basis, a CNN-BiGRU-Transformer model is constructed, which utilizes CNN-BiGRU to extract spatiotemporal features and dynamic weight allocation of Transformer. Finally, the crested porcupine optimization algorithm is used to adaptively search for BiGRU model parameters, including training epochs, number of hidden layer neurons, and learning rate, in order to achieve automatic optimization and performance improvement of the CNN-BiGRU-Transformer model structure. The simulation results of the integrated energy system of cooling, heating and power show that the proposed model achieves higher prediction accuracy and stabler error distribution compared to existing load forecasting models.