Electricity load forecasting method based on SCI-MN-SE model
Aiming at the problems of low extraction efficiency of time series information and low accuracy of prediction results in existing power load forecasting methods,a power load forecasting model(SCI-MN-SE)based on Sample Convolution and Interaction Network(SCINet)and Multi-Layer Perceptron(MLP)is proposed.Feature extraction of load data at different temporal resolutions is performed using SCINet to better capture the relationship between short-term and long-term information in temporal features;The extraction efficiency of time series information is improved by a bottleneck module consisting of a superposition of multilayer perceptron machines,which performs a hybrid operation along the temporal and feature dimensions.The weighting of the effective information is improved by combining channel attention to enhance the prediction accuracy.The experimental results on the real power load dataset of a region in Australia show that the average absolute percentage error is reduced by 26.30%and 24.27%compared with TCN,Informer and other methods,respectively,which proves that the SCI-MN-SE model can extract the time series information efficiently and the prediction effect is better.
electricity load forecastingmultilayer perceptronsample convolution and interactionchannel attention
万方数据
维普
