Short-Term Power Load Forecasting Based on Neural Network VMD_CNN_BiGRU
Shuaiying You, Xiangxiang Chen, Ke Xi, Min Chen · IEEE Access · 2025
In response to the challenges posed by large data fluctuations, complex influencing factors, and the need for high forecasting accuracy in short-term power load forecasting, this paper proposes a combined convolutional neural network-bidirectional gated loop unit forecasting model based on variational modal decomposition. Firstly, Pearson’s correlation coefficient method is employed to identify the key factors influencing the power load. Secondly, the variational modal decomposition (VMD) technique is utilized to decompose the original data into multiple smooth intrinsic modal functions (IMFs). Lastly, the convolutional neural network (CNN) is integrated with the bidirectional gated recurrent units (BiGRUs), resulting in the construction of a hybrid model, designated as VMD_CNN_BIGRU. The results demonstrate that the model effectively leverages the data decomposition capabilities of the VMD algorithm and the strengths of the CNN_BiGRU model in feature extraction and time series prediction. This integration significantly enhances the accuracy of load forecasting. The experimental outcomes illustrate that the hybrid model exhibits an improvement of 24% in the R metric compared to the individual models. Meanwhile, the model obtained a mean absolute percentage error (MAPE) of 1.77% in the public dataset, an root mean square error (RMSE) value of 18429.00 MW (MEGAWATT), a mean absolute error (MAE) value of 12292.74 MW, and an R-squared value of 0.94.