基于高斯过程的短期风电功率概率预测

摘要

风电是近年来发展迅速的绿色能源,因此对短期风电功率的预测就显得尤为重要.提出了基于高斯过程(Gaussian Processes,GP)的概率预测方法,详细说明了该方法的短期风电功率概率预测原理并建立了数学模型.为了使短期风电功率概率预测精度有一个良好的对比性分析,将基于不同的单一协方差函数以及组合协方差函数的GP方法用于预测中,以国外某风电场2006年6月份的历史风电功率实测数据进行算例实例分析,并与SVM方法在同等条件下进行比较.实验结果表明,GP方法均可以给出较好的预测效果,优于SVM的预测结果,且能给出预测输出的置信水平.若考虑具有自动相关确定(Automatic Relevance Determination,ARD)协方差函数或具有ARD特性的组合协方差函数时,GP方法的预测效果最好.%Wind power is a green energy developed rapidly in recent years,so the short-term wind power prediction is particularly important.A short-term wind power probability prediction method based on Gaussian Processes (GP) is presented.The principle of the method is decribed and a mathematical model is established.In order to make the prediction accuracy has a good comparative analysis,a GP method based on different single covariance function and combined covariance function was used in the prediction.A case study of historical wind power measured data of a wind farm in a foreign country in June 2006 was carried out and compared with SVM method under the same conditions.The experimental results show that the GP method has better prediction accuracy than SVM,and can give the confidence level of the predictive output.The GP method has the best predictive performance when using the automatic relevance determination (ARD) covariance function or the combined covariance function with ARD characteristic.