Monthly Wind Power Forecasting: Integrated Model Based on Grey Model and Machine Learning

Wind power generation has been developed rapidly due to rising global interest in renewable clean energy sources. Accurate prediction of the potential amount of such energy is of great significance to energy development. As wind changes greatly by season, time series analysis is considered as a natural approach to characterize the seasonal fluctuation and exponential growth. In this paper, a dual integrated hybrid model is presented by using random forest (RF) to incorporate the extreme gradient boosting (XGB) with empirical mode decomposition (EMD) and a fractional order accumulation seasonal grey model (FSGM). For seasonal fluctuation in vertical dimension processing, the time series is decomposed into high and low frequency components. Then, high and low frequency components are predicted by XGB and extreme learning machine (ELM), respectively. For the exponential growth in horizontal dimension processing, the FSGM is applied in the same month in different years. Consequently, the proposed model can not only be used to capture the exponential growth trend but also investigate the complex high-frequency variation. To validate the model, it is applied to analyze the characteristics of wind power time series for China from 2010 to 2020, and the analysis results from the model are compared with popularly known models; the results illustrate that the proposed model is superior to other models in examining the characteristics of the wind power time series.

Automated summary

In this paper, a dual integrated hybrid model is presented for accurate prediction of the potential amount of wind power generation by incorporating random forest (RF) with extreme gradient boosting (XGB), empirical mode decomposition (EMD), and a fractional order accumulation seasonal grey model (FSGM).