Multi-step Wind Speed Forecasting Based On Gated Recurrent Unit And Kernel Density Estimation

Wind energy,as a clean and low-carbon renewable energy,has been widely used in the whole country and even the world.Wind speed,as a major factor affecting wind energy,has characteristics of volatility,non-linearity,uncertainty and seasonality,which bring severe challenges to wind power grid.Accurate forecasting of wind speed is the key to solve this problem.Considering the complex characteristics of wind speed,a point forecasting model based on gated recurrent unit and an interval forecasting model based on nonparametric kernel density estimation are established in this paper to realize the effective forecasting of wind speed.The main contents are as follows:(1)Taking the actual wind speed data collected from a wind farm in Inner Mongolia as the research object,the quartering method is adopted to identify and eliminate the abnormal data in the wind speed data set,and the interpolation is carried out according to the missing data categories.The wind speed sequence of four seasons is decomposed and reconstructed by the singular spectrum analysis and variational modal decomposition algorithm with parameter optimization which made it have more obvious characteristics and laid a foundation for the establishment of wind speed forecasting model.(2)Aiming at each wind speed sequence after pretreatment,a wind speed forecasting model based on gated recurrent unit was established to realize the multi-step rolling forecasting of wind speed.The experimental results show that the forecasting model based on the gated recurrent unit can effectively track the changing trend of wind speed signals and has a strong ability to learn the irregular trend of wind speed.The preprocessing and decomposition-reconstruction of wind speed series are beneficial to reduce the complexity of wind speed time series and improve the forecasting effect effectively.The bidirectional gated recurrent unit has more advantages in acquiring information and learning the variation trend of wind speed signal.At the same time,the seasonal adaptability of the model is verified.(3)According to the error distribution characteristics of wind speed series training set,a wind speed forecasting model based on Gaussian kernel and optimal window width kernel density estimation is proposed.Combined with the error cumulative distribution function,the error fluctuation ranges under different confidence degrees are calculated.The experimental results show that the wind speed forecasting model based on Gaussian kernel and optimal window width kernel density estimation has high fitting accuracy and seasonal adaptability.When the confidence level is around 90%,the interval width is relatively narrow and the interval coverage rate is relatively high,so the forecasting interval at this time is the most reasonable.