Short-term forecasting for multiple wind farms based on transformer model

Abstract

With the rapid growth of wind power installed capacity in recent years, the distribution of wind farms will be relatively dense, and there are usually multiple wind farms in the same area. However, due to the complex correlations and dependencies among these wind farms, traditional forecasting models for individual wind power are difficult to apply. Meanwhile, the accurate forecasting of power output of multiple wind farms is very important to the evaluation results of the renewable energy consumption capacity of the grid, and this problem has received extensive attention from many scholars. To improve the accuracy of forecasting, we apply the Transformer model from natural language processing (NLP) to the field of wind power forecasting. The proposed model is capable of capturing longer sequence internal dependencies, as well as capturing the key information of wind data in a comprehensive and multifaceted way. By comparing with the comparison method, case studies show that the model is not only able to accurately extract different levels of correlation between multiple wind farms, but also to give accurate wind power forecasting results.