Multistep power load forecasting using iterative neural network-based prediction intervals

Abstract

Our study investigates the multistep forecasting problem under power load fluctuation. This paper proposes a multistep prediction interval (PI) construction approach that is developed using an iterative neural network (NN) to handle data uncertainty. First, a new data processing method is extended that combines variational modal decomposition (VMD) and partial autocorrelation analysis (PACF). The iterative lower upper bound estimation (ILUBE) model is developed that uses an NN with interval input, output, and feedback iteration. It is necessary to formulate a new performance index and adapt it to the requirements of multistep PIs by introducing a prediction horizon factor and a prediction deviation value. Finally, a particle swarm optimization (PSO) algorithm is employed to search for the optimal parameters of the ILUBE. Comparative experiments are carried out using the power load data from two regions in a case study. The results indicate that the proposed approach increases the quality of PIs and the prediction horizon.