An improved method for predicting recurrence period wind speed considering wind direction

In light of extreme value distribution probability, an improved prediction method of the Recurrence Period Wind Speed (RPWS) is constructed considering wind direction, with the Equivalent Independent Wind Direction Number (EIWDN) introduced as a parameter variable. Firstly, taking the RPWS prediction of Beijing city as an example, the traditional Cook method is used to predict the RPWS of each wind direction based on the measured wind speed data in Beijing area. On basis of the results, the empirical formulae to determine the parameter variables are fitted to construct an improved expression of the non-exceedance probability of the RPWS. In this process, the statistical model of the optimal threshold is established, and thus the independent wind speed samples exceeding the threshold are extracted and fitted to follow the Generalized Pareto Distribution (GPD) model for analysis. In addition, the Extreme Value Type I (EVT I) distribution model is used to predict and analyze the RPWS. To verify its wide applicability, the improved method is further used in cities like Jinan, Nanjing, Wuxi, Shanghai and Shenzhen to predict and analyze the RPWS of each wind direction, and the prediction results are compared against those gained via the traditional Cook method and the whole direction. Results show that the 50-year RPWS results predicted by the improved method are basically consistent with those predicted by the traditional method, and the RPWS prediction values of most wind directions are within the envelope range of the whole wind direction prediction value. Compared with the traditional method, the improved method can readily predict the RPWS under different return periods through empirical formulae, and avoid the repeated operation process and some assumptions in the traditional Cook method, and then improve the efficiency of prediction. In addition, the improved RPWS prediction results corresponding to the GPD model are slightly larger than those of the EVT I distribution model.