Short-term power load interval prediction based on LSTM&CNN-SVRKDE

Aiming at the problems that the single-point forecasting model is difficult to describe the load change trend more comprehensively, and the optimization algorithm has the problems of easy local convergence and poor global search ability, an IFWA-LSTM load forecasting model based on the Improved Fireworks Algorithm (IFWA) to optimize the parameters of the Long and Short-Term Memory Neural Network (LSTM) model was proposed. The global search capability and flexibility of the fireworks algorithm (FWA) were improved by introducing an adaptive adjustment factor and an elite strategy to enhance the solution efficiency and accuracy. At the same time, Kernel Density Estimation (KDE) was used to probabilistically predict the prediction results obtained from the IFWA-LSTM model, and for the problem of weak nonlinear mapping ability of the KDE model, Support Vector Regression (SVR) was used to optimally fit it, and then the probability density function of the predicted values was solved to obtain the power prediction intervals under different confidence levels. Based on the measured load data of a station area, the proposed model was verified experimentally. The results show that compared with the traditional neural network algorithm model, the proposed model has higher prediction accuracy, smaller result error, and better overall prediction performance.