Short-term electric load forecasting method based on hybrid ensemble deep learning framework

Short-term electric load forecasting is crucial for the safe and economical operation of distribution networks. Due to the non-stationary and nonlinear variations of load caused by weather and periodicity, the accuracy of a single prediction model often fails to meet practical requirements. This paper proposes a hybrid prediction model based on ensemble learning, integrating least squares support vector regression (LSSVR) and convolutional neural network-long short-term memory (CNN-LSTM) models. Firstly, the seasonal decomposition algorithm is performed to divide the electric load into periodic load components and non-periodic load components. Subsequently, based on the bagging ensemble learning strategy, two types of deep prediction models are build to predict the non-periodic load components separately. Finally, the composite prediction value is obtained through weighted averaging. Simulation experiments on real load datasets verify the effectiveness of the proposed hybrid prediction model