The Short-Term Electrical Load Forecasting Based on Gated Recurrent Unit Combination

Short-term electrical load forecasting is the key technology to ensure the safety and economic operation of electrical grid. However, the traditional model faces the problems of mode aliasing, noise residue and low calculation efficiency. Therefore, this paper proposes a combined forecasting model (EGM) based on Complementary Ensemble Empirical Mode Decomposition, Gated Recurrent Unit and Multiple Linear Regression. Firstly, use CEEMD to decompose the electrical load signal, and the modal aliasing is suppressed by using the paired positive and negative white noise cancellation mechanism. Secondly, a hierarchical prediction framework is constructed. GRU neural network is used to capture complex time series characteristics for high-frequency components, and MLR is used to quickly fit periodic laws for low-frequency components. Finally, high-precision load forecasting is realized through the reconstruction of component forecasting results. Compared with eight benchmark models, the results show that the mean absolute percentage error and root mean square error of EGM are significantly better than other methods, and it is more advantageous in calculation time. This model provides a solution with both theoretical innovation and engineering practicability for load forecasting of new electrical system, and can effectively support dynamic dispatching and equipment optimization of distribution network.