An integrated prediction model based on meta ensemble learning for short-term wind speed forecasting

Abstract

As wind power increasingly integrates into power grids and energy systems, accurate and reliable wind speed forecasting (WSF) has become essential for wind power scheduling and management. Considering the fluctuating and random characteristics of wind speed, a novel integrated model for short-term WSF is developed in this work, which integrates multiple models through meta ensemble learning to achieve better generalization, robustness, and accuracy. This model consists of four components: data input, base predictor, meta ensemble learning, and prediction data output. The base predictor component includes multiple pre-trained base predictors of long short-term memory recurrent neural network to provide initial prediction values for wind speed. The meta ensemble learning component is a multi-input and multi-output back propagation neural network that outputs automatically adjust weight coefficients for various base predictors, drawing on the environmental and meteorological characteristics of historical wind speed data. The ultimate prediction result of wind speed is obtained through a weighted summation of the initial prediction values of base predictors. The authors assess the effectiveness of the integrated WSF model by contrasting its performance with that of alternative forecasting models. The simulation results reveal that the proposed integrated prediction model surpasses both individual prediction models and traditional integrated prediction approaches in terms of prediction stability and accuracy for short-term WSF.