Statistical Forecasting of Short-term Electricity Load Using a Meteorological Feature-driven Deep Learning Fusion Framework

It is crucial to precisely forecast the power load to achieve efficient resource distribution and cost control. The current work introduces a deep learning (DL) driven fusion framework for short-term load forecasting (STLF). The study indicates that many factors, including meteorological conditions, significantly influence the power load. In this paper, the convolutional neural network (CNN), long short-term memory network (LSTM), and gated recurrent unit (GRU) are introduced to establish the CNN-GRU and CNN-LSTM models to improve the prediction accuracy. These models combine the CNN's feature extraction power and the recurrent neural network (RNN) advantages in time series modeling. In the empirical analysis, the power load and meteorological data from 2021 to 2023 in Pingxiang City, Jiangxi Province, China, are employed to assess the model’s efficiency. The experimental outcomes indicate that the CNN-GRU model exhibits the best performance according to the root mean square error (RMSE), mean absolute error (MAE) and other indicators while providing a forecasting effect superior to other models. Furthermore, the 5-fold cross-validation is employed to further verify the model stability and generalization ability.