A multi-energy load forecasting method based on parallel architecture CNN-GRU and transfer learning for data deficient integrated energy systems

In the integrated energy system with small samples, insufficient data limits the accuracy of energy load forecasting and thereafter affects the system's economic operation and optimal dispatch. For this specific environment, this paper proposes a multi-energy load forecasting method based on the neural network model and transfer learning to meet the demand of enterprises for forecasting accuracy. The method improves forecasting accuracy through three stages including data analysis and processing, a combined model built and load forecasting. More specifically, the Pearson correlation coefficient is used to filter out meteorological variables with strong correlation based on energy load and meteorological data. A combined model is developed based on the convolutional neural network and gated recurrent unit. A model structure adjustment strategy based on the maximum mean difference is proposed to dynamize the structure to cope with the complex prediction environment. The synergy between source and target domain data is realized based on transfer learning. In addition, the model performance is further optimized through model training, transfer learning, and parameter fine-tuning, which lays the foundation for improving the forecasting accuracy of electricity, gas, cooling, and heating loads. The simulation results show that the proposed method can achieve satisfactory predictions for integrated energy systems with small sample data.