Study On Probabilistic Prediction Of Wind Power In Wind Farm

As the world's largest consumer of energy, the high dependence on oil and gas energy leads that energy security issues, environmental issues have become increasingly obvious in China. The continuous improvement of wind power technology and the increasing of wind farm installed capacity and wind power grid connected scale bring our country the huge economic benefits, meanwhile, ease the energy security and bring environmental benefits. In order to ensure the stable operation of the power system and the reliability of the power supply system, the power supply system must be effectively planned and scheduled. However, as a kind of intermittent power supply, the volatility and uncertainty of wind power, increase the difficulty of scheduling of grid security and the burden of system reserve capacity. Therefore, the accurate prediction of the output power of the wind farm can not only benefit to reasonably arrange the scheduling plan to achieve the balance of supply and demand of the power system, but also can effectively reduce the spare capacity and reduce the total cost of the power generation.In this paper, the domestic and international situation of wind farm wind power prediction systems are introduced. Among them, most of the prediction systems use single prediction, these methods can display the maximum possible power value in the future period. However, single prediction methods can not provide the uncertain information of wind power output of the wind farm. The emergence of the probabilistic forecasting methods is the key to solve the lack of the uncertainty information of wind power prediction. These methods can not only display expectations as the single point values predicted results, but also display probability distribution as the prediction error distribution information which provide more abundant information for power system dispatching decision-making and operation risk assessment.Starting from these two kinds of single forecasting methods:ARMA and SVM, the characteristics of these two single valued prediction methods have been analyzed. Then, using ARMA and SVM as the function model, the probability prediction models of ARMA-quantile and SVM-quantile are built. In the process of analysing the SVM-quantile probability prediction method, the probability density function of the quantile method is obtained, at the same time, puts forward subsection modeling method with regard to the subsections whose test results are not passed. Then, the probabilistic prediction method based on relevance vector machine is studied, in the process of improving the RVM probability prediction method, according to the error sequence from the two kinds of single forecasting method of ARMA and SVM, a probabilistic forecasting method based on prediction error is put forward. Finally, using the same evaluation criteria, the two kinds of probability prediction methods based on quantile regression and relevance vector machine are compared in single valued prediction results, probabilistic forecasting results and running time cost.