Research On Short-term Wind Power Range Forecasting Based On Physical Model And Data-drive

With the increasing emphasis on ecological environment protection worldwide,developing renewable energy is an important way to build a low-carbon economy.As a renewable energy industry,wind power generation has been rapidly developing worldwide.However,the uncertainty of wind power brings great challenges to the power grid system,and more accurate wind power forecasting methods are needed to ensure the safe and stable operation of wind power integration.Currently,with the continuous expansion of the wind power industry,the demand for accuracy of wind speed data is increasing,and accurate wind speed forecast data has become an important factor in ensuring the stable and safe operation of the power grid.Artificial intelligence methods are widely used in wind power forecasting research,but the selection of model parameters has an important impact on the model performance.It is necessary to select an appropriate combination of parameters during modeling to fully exploit the model’s performance.Since the output power has uncertainty,errors in wind power point predictions are inevitable.Therefore,it is necessary to conduct power interval predictions based on power single point predictions to more effectively support the power dispatch department in making dispatch decisions and conducting system reliability analysis.This paper conducts in-depth research from the perspectives of wind speed data,optimization of forecasting model parameters,and interval prediction.The specific content is as follows:(1)Due to the inherent errors in wind speed forecasts from the WRF model,directly using them can greatly affect the accuracy of wind power prediction.Moreover,considering that the target wind farm is located offshore and the summer and autumn seasons are prone to typhoons and tropical cyclones,this paper proposes a wind speed ensemble forecasting model based on WRF model forecasts to reduce wind speed forecast errors.Firstly,multiple parameter schemes based on cumulus clouds are used to forecast the wind speed of the target wind farm.Then,a wind speed ensemble forecasting model is constructed by combining Bi LSTM,CNN,and attention mechanisms(AM)based on the forecasted wind speed data.Furthermore,the ensemble forecasted wind speed is subjected to secondary correction.Experimental results show that the proposed method effectively reduces wind speed forecast errors and improves the accuracy of numerical weather forecast wind speed.The wind speed results after the secondary correction will be applied to wind power point prediction.(2)Due to the limitations of using a single model for wind power forecasting,the optimal prediction performance cannot be achieved.Therefore,this paper proposes an improved aquila optimizer(IAO)that enhances the global and local search process of the aquila optimizer(AO),and combines it with the least squares support vector machine(LSSVM)to establish the IAOLSSVM composite power prediction model.The improved aquila optimizer is used to optimize two important parameters of the LSSVM,resulting in an optimal prediction model that improves prediction accuracy.This composite model is used in conjunction with the ensemble forecast wind speed secondary correction results for short-term wind power forecasting.Experimental results demonstrate that the IAO-LSSVM composite model has better predictive performance.(3)Based on the single-point prediction of the IAO-LSSVM combined model,in order to provide more comprehensive support for power dispatch departments to make scheduling decisions and analyze the reliability of the power grid system,this paper uses the kernel density estimation method(NKDE)for short-term power interval prediction.The data reflection method is introduced to correct the boundary error of the kernel density estimation method.To improve the local adaptability of the method,an adaptive bandwidth is used to replace the fixed bandwidth of the kernel density estimation,and the power error data is fitted with a probability density function to obtain the power fluctuation interval under a specific confidence level.Experimental results show that the proposed method has higher coverage rate and narrower interval width,and better overall performance.