Research On Short-term Wind Power Forecasting Method

In recent years,with the further adjustment of China's energy structure system,wind power industry has been developed rapidly with the support of national policies,and the scale of wind power network has gradually expanded.However,with the characteristics of intermittent,volatility and randomness of wind power,the expansion of wind power network has brought difficulty to power dispatching,which has seriously affected the safety,reliability,economy and stable operation of power grid.Accurate and efficient prediction of wind farm output can effectively alleviate the difficulty of power dispatching and reduce the threat of wind power network.Therefore,it has practical engineering value and significance to forecast the output of wind farm.This paper mainly studies the wind power forecast as follows:(1)In this paper,the least Squares support vector machine theory is used to construct a deterministic short term wind power forecasting model.Because the prediction performance of the LSSVM model is influenced to a great extent by the choice of super parameters and kernel function,four kinds of common kernel functions are selected to construct the LSSVM prediction model,and an improved gravitational search algorithm is proposed to select the parameters of LSSVM predictive model,and establish a short term wind power forecasting model for IGSA optimization LSSVM.The simulation results show that the wind power prediction performance of the IGSA-LSSVM model with RBF kernel function is superior to other kernel functions,and the Igsa optimization LSSVM method has better stability and higher accuracy than other intelligent algorithm optimization LSSVM.(2)Due to the foresight the deterministic prediction results of wind power have more or less inevitable deviations,in order to obtain the uncertain information of wind power prediction results,the error analysis and statistics of deterministic single point prediction are obtained based on the single point prediction,and the nonparametric kernel density estimation interval prediction method is used,The fluctuation of wind power under different confidence levels is presented based on the statistical characteristics of the error.The simulation results show that the prediction method based on nonparametric kernel density estimation can be used to estimate the future fluctuation of wind power,and it has a reference value for engineering application.(3)Considering that the interval prediction method based on nonparametric kernel density estimation needs to be a deterministic point prediction in order to obtain the future wind power prediction interval,a large number of power prediction error data needs to be analyzed and processed statistically.An empirical modal decomposition based on complementary set complete ensemble empirical mode decomposition and improved gaussian process regression,the short term wind power probability forecasting method,firstly,the wind power is decomposed by CEEMD to obtain the wind power sequence,the forecasting model is established by the IGSA and the multi-kernel covariance function is applied to the GPR model.The final probability prediction results are obtained by the wind power sequence.The simulation results show that the prediction model based on CEEMD and improved GPR can make accurate and effective probability forecast,has better probability evaluation index,and can get better single point prediction value and confidence interval,and the prediction result is good.