Power Load Forecasting Method Based on Random Matrix Theory and CNN-LSTM Model

Abstract

Rapid and accurate load forecasting is the premise of economic operation of comprehensive energy system. A short-term load forecasting method based on random matrix theory and CNN-LSTM model was proposed to solve the problem of complex coupling relationship and strong load fluctuation in integrated energy system. Firstly, the high-dimensional random matrix is constructed and the coupling characteristic matrix is calculated, and the coupling relation of each characteristic quantity is extracted from the time dimension. Then, the coupling feature matrix is compressed and enhanced based on one-dimensional convolutional neural network to extract the coupling features. Finally, load prediction of coupled data is carried out based on long and short term memory network model. In this paper, the load data of a building is used as the data source for simulation analysis, and the results of an example prove the correctness and effectiveness of the proposed prediction method.