Research on Short-Term Power Load Forecasting Method Based on PCA-VMD-LSTM-MTL

Based on the standard back-propagation neural network, the paper proposes a method combining principal component analysis, variational mode decomposition, improved error back-propagation neural network and multi-task learning to predict the medium and short term power load of the power grid. Firstly, the principal component analysis method is used to extract the characteristics of the influencing factors of the power load, effectively reducing the dimensions of the data samples, eliminating the redundancy and linear information of the data, and retaining the main components as the input data of the model. Then, a combined prediction model based on the combination of variational modal decomposition (VMD), short-term and short-term memory network (LSTM) and multi-task learning (MTL) is proposed. VMD is used to decompose the original load sample set into several different eigenmode functions (IMF) and a residual component to reduce the nonstationarity and complexity of the load data sample set. LSTM model is used to predict the decomposed multi-modal components separately, and MTL is used to establish a shared learning layer to speed up the operation efficiency, and the prediction results of different components are superimposed to obtain the final load prediction value. Finally, through simulation experiments, the prediction results of PCA-VMD-LSTM-MTL, BP model, ELM model and LSTM show that the prediction effect of PCA-VMD-LSTM-MTL model is better than the other three prediction models, and it shows good performance in short-term power load forecasting.