Wind speed forecasting using multi-scale feature adaptive extraction ensemble model with error regression correction

Jun Chen

doi:10.1016/j.eswa.2022.117358

Return

Wind speed forecasting using multi-scale feature adaptive extraction ensemble model with error regression correction

2022-11-01

PRE

DOI：10.1016/j.eswa.2022.117358

Original

Original request for help

Save

Abstract

En 中文

Wind speed forecasting is vital for the operational safety and power generation quality of wind power systems. However, owing to the strong non-stationary and nonlinear characteristics of natural wind, high performance wind speed forecasting is still a challenging task. In this study, a novel multi-scale feature adaptive extraction (MSFAE) ensemble model is developed to provide accurate and reliable wind speed forecasting. The proposed model constructs six GWO-CNN-BiLSTM (GCNBiL) networks with different lengths of convolution operators, and extracts and learns the deep autocorrelation feature hidden in high-resolution data at different time scales. Subsequently, the proposed model employs a multi-objective cuckoo-search-moth-flame hybrid optimization (MOCSMFHO) algorithm to ensemble the forecasting results of six GCNBiL benchmark models. Finally, the model uses a multi-objective error regression method for error correction. Three actual wind series collected from Xinjiang, China, were selected to verify the proposed model. These case studies indicate that the proposed model outperforms all sixteen alternative models as well as the three state-of-the-art models with a 40% average improvement ratio. Moreover, the developed GCNBiL models have excellent prediction accuracy, resist overfitting, and show an 18% improvement in forecasting performance. Furthermore, the proposed MOCSMFHO algorithm possess both excellent global search accuracy and convergence speed, and can effectively integrate the prediction advantages of autocorrelation features in different time scales to obtain a 4% improvement in accuracy. The proposed multi-objective error regression correction method can also provide more than 26% improvement in prediction performance and delivers better error correction performance than the existing methods in practical applications.

Keywords:

Wind speed forecasting

Multi -scale feature adaptive extraction

Ensemble model

Error correction