基于分位数回归的改进权重GRU风电功率区间预测  

IMPROVED WEIGHTED GRU WIND POWER INTERVAL PREDICTION BASED ON QUANTILE REGRESSION

作　　者：柳天虹 齐胜利 裔扬[1] 菅利彬[2] 乔显著 章恩泽 Liu Tianhong;Qi Shengli;Yi Yang;Jian Libin;Qiao Xianzhu;Zhang Enze(School of Information Engineering,Yangzhou University,Yangzhou 225127,China;NARI Technology Co.,Ltd.,Nanjing 211106,China)

机构地区：[1]扬州大学信息工程学院,扬州225127 [2]国电南瑞科技股份有限公司,南京211106

出　　处：《太阳能学报》2024年第12期291-298,共8页Acta Energiae Solaris Sinica

基　　金：国家自然科学基金(62203381);江苏省自然科学基金(BK20190876)。

摘　　要：为提高短期风电功率预测的准确性,解决点预测难以描述风电功率不确定性,且在数据发生突变时传统GRU无法准确跟踪数据突变问题,提出一种基于分位数回归的改进权重GRU风电功率区间预测模型(QR-EGRU)。首先采用改进的自适应小波阈值去噪方法对原始数据降噪处理,减少数据噪声影响;然后引入两个更新门权重矩阵代替传统更新门权重矩阵,新的权重矩阵采用信息熵动态调整矩阵的变化趋势,量化权重的变化程度,构建信息熵权重门控循环单元(EGRU)网络;最后基于分位数回归算法获取不同分位数下的点预测概率区间。通过风电场的有功功率进行实验验证,结果表明:相比于其他对比方法,所提出的模型在相同实验条件下能提高预测精度,具有较好的风电功率区间预测性能。In order to improve the accuracy of short-term wind power prediction,an improved entropy weighted GRU wind power interval prediction model based on quantile regression(QR-EGRU)is proposed to solve the problems that point prediction is difficult to describe wind power uncertainty and the traditional GRU cannot accurately track the data mutation when data changes.Firstly,the improved adaptive wavelet threshold denoising method is used to reduce the noise of the original data.Then two update gate weight matrices are introduced to replace the traditional update gate weight matrix.The new weight matrices adopt information entropy to dynamically adjust the matrix changes,quantify the change degree of the weights,and construct the information entropy weighted GRU(EGRU)network.Finally,the probability interval under different quantiles of the point prediction is obtained based on quantile regression algorithm.Experimental results show that the proposed model can improve the prediction accuracy compared with other comparison methods under the same experimental conditions and has a better interval prediction performance.

关 键 词：风电功率 预测 门控循环单元 分位数回归 信息熵 小波阈值去噪 

分 类 号：TM614[电气工程—电力系统及自动化] TP18[自动化与计算机技术—控制理论与控制工程]