Weather Research Forecasting Model Integrated Apriori Algorithm For Short-term Wind Speed Forecasting

The increasingly rapid development of society needs huge power consumption. Currently, energy poverty has been an unavoidable problem faced by all the countries worldwide. Renewable energy, especially wind energy, gradually becomes a new concern to address the global energy crisis and environmental deterioration. Being a kernel study in wind power generation, wind speed prediction is a significant and tough task, which this paper puts effort on.In the circumstance of predicting short-term wind speed (48-72h ahead), the methods of numerical weather prediction are usually employed. However, the initial disturbance and the model uncertainty of the physical process are the most important factors that constrained the forecasting accuracy to be improved. The main idea of this paper is to propose a new post-processing optimization approach for the numerical weather prediction models. Weather Research Forecasting model, one of the most widely used physical models, is chosen to forecast the following72hours wind speed series of the wind field. Then decompose the forecasting series by using the Spatial Box filter decomposition. According to the stagnation points, a72-h forecasting sequence can be divided into several waves or stages. And find out the association rule among the relative length, relative height of a wave and its relative error by using the Apriori algorithm. Finally, estimate the modified weight of each wave according to the relative error, and multiply the wave by the modified weight. The post-processing optimization approach consists of four steps:(Ⅰ) Spatial Box filter decomposition;(Ⅱ) clustering;(Ⅲ) mining of the association rule by using Apriori algorithm;(Ⅳ) optimization. Wind speed data of Chengteh and Ulanqab, China, is used for validity analysis. The simulation results show that:(1) the Apriori algorithm can discovered association rules between the forecasting wind speed series and its relative error;(2) the post-processing optimization approach can improve the forecasting accuracy significantly; and (3) the methodology do well performance on different simulation location outputs to the leading elimination of systematic errors.