부산대학교 도서관

Aiming at the problems of short-term power load’s strong randomness and low forecasting accuracy, a combination forecasting method based on variational mode decomposition (VMD) and gate recurrent unit (GRU) neural network is proposed. This method uses VMD technology to decompose the original load sequence into sub-sequences with different characteristic frequencies, and establishes a forecasting model for each sub-sequence. The load forecasting model uses GRU neural network. Hyperparameter optimization based on genetic algorithm (GA) makes it not only has good local search ability under different parameters, but also strengthens global search ability. Experiments show that the model has better regression accuracy and generalization ability, it can get more accurate forecasting results.