Load Prediction Based on Depthwise Separable Convolution Model

In recent years, deep learning has been widely used in load forecasting. While dealing with load forecasting problems, the problems of large-scale calculation and parameter redundancy are made by ordinary convolutional networks, which leads to large network storage space and long calculation time. Thus, a lightweight convolutional network structure by substituting depthwise separable convolution (DSC) for ordinary convolutions in convolutional networks was designed in this paper. First, the ordinary convolution is separated in the spatial dimension through channel convolution to increase the network width and expand the feature extraction range, and then use the point-by-point convolution to reduce the computational complexity of the ordinary convolution operation. The given load data is preprocessed before establishing the forecast model. Firstly, the collected data are extracted and analyzed for the load characteristics, and then the correlation between the load data and the factors that affect the load value is analyzed to determine the main factors. For the load data obtained after preprocessing, a DSC neural network is used to establish a load prediction model. The load forecasting results show that the proposed method is effective and correct.