基于时域卷积网络和分位数回归的短期风电功率概率预测

Probabilistic Power Prediction of Short-Term Wind Generation Based on Temporal Convolutional Network and Quantile Regression

提出一种基于时域神经网络结合分位数回归与动态误差修正的风电功率概率预测方法。首先,采用时域卷积网络构造分位数回归模型。然后,对分位数回归模型的预测误差分布进行建模。最后,采用动态误差修正技术对分位数点进行修正,得到多组风电功率预测区间。采用两个风电场数据集进行验证,结果表明所提方法能在保证高可靠性的同时获得较窄的预测区间,并有效解决分位数交叉问题,可为电网调度优化提供有效帮助。

A power probabilistic power prediction method for wind generation based on the combination of temporal neural network,quantile regression and dynamic conformalization is proposed in this paper.First a quantile regression model is constructed using temporal convolutional networks.Then the prediction error distribution of quantile regression is modeled.Finally the quantile points are corrected by dynamic error correction technology and multiple groups of wind power prediction intervals are obtained.Two wind farm datasets are used for verification.The results show that the proposed method can obtain a narrow prediction interval while ensuring high reliability,and effectively solve the quantile crossing problem,which can effectively help in optimizing power grid dispatching.