基于RBF神经网络的风电场风速区间预测

摘要

风电场风速的预测受多种因素影响，且容易引入噪声信息，具有较强的不确定性。因此有必要研究其预测区间，从而描述和量化这种不确定性。现有的区间预测方法往往覆盖概率过低或者依赖于初始值的选取。提出了一种基于RBF神经网络构建预测区间的方法，根据点预测的输出值与实际观测值的残差估计训练样本预测区间的上界和下界，初始化模型参数，然后以基于覆盖概率和区间宽度的综合准则作为目标函数更新权值。分别采用江苏、宁夏和云南的风电场风速数据进行实验，结果表明，该方法覆盖概率PICP及综合评价指标CWC优于LUEM与LUBE这两种经典的区间预测方法。%In this paper,a new method is proposed for the construction of PI based on RBF neural network.The residuals between the model outputs and the corresponding observed data are used for estimating the lower and upper prediction limit of training samples.The parameters of neural network are initialized and then trained by a PI-based objective function, which involves both coverage probability and interval width.Experiments are conducted with wind speed data sets of Jiang-su,Ningxia and Yunnan.