基于DMD-NARX模型的短期电力负荷预测方法  

Short-term power load forecasting method based on DMD-NARX model

作　　者：王雪菲 李勇[2] 余国晓 程晨[1] 杨辉军[1] WANG Xuefei;LI Yong;YU Guoxiao;CHENG Chen;YANG Huijun(School of Information Engineering,Anhui Institute of International Business,Hefei 231131,China;Institute of Statistics and Applied Mathematics,Anhui University of Finance&Economics,Bengbu 233030,China;School of Electronics and Information Engineering,Anhui University,Hefei 230601,China)

机构地区：[1]安徽国际商务职业学院信息工程学院,合肥231131 [2]安徽财经大学统计与应用数学学院,蚌埠233030 [3]安徽大学电子信息工程学院,合肥230601

出　　处：《黑龙江大学自然科学学报》2022年第3期307-316,共10页Journal of Natural Science of Heilongjiang University

基　　金：国家自然科学基金资助项目(61672032);安徽高校自然科学研究资助项目(KJ2021A1529)。

摘　　要：提出了一种基于DMD-NARX模型的短期电力负荷预测方法,深入地探索了负荷变化趋势和历史数据之间的内在关联,同时在短期预测的精度上有所提高。首先通过自相关函数(Autocorrelation function, ACF)并结合短期负荷波动的时间规律特性,在已有历史相关数据的基础上推导出相应日期的输入特征集合;然后将输入特征集合归一化后通过Hankel矩阵完成由单变量输入特征序列向多维数据矩阵的转换,以动态模态分解(Dynamic mode decomposition, DMD)为手段完成对上一步所得多维数据矩阵的动态模态估计和特征分解,同时对电力负荷底层的多尺度动态情况有了更加深入的掌握;最后使用基于外部输入的非线性自回归(Nonlinear autoregressive with external inputs, NARX)神经网络模型,同时以上一步取得的动态模态估值作为计算相应预测日期内各时段负荷分布的基础,并推导出最终预测结果。最终的测试数据证明,此方法较好地改善了模型的预测精度。In order to excavate the potential relationship between the changing trend of load series and the historical data,and improve the accuracy of the short-term load forecasting,a short-term load forecasting method based on DMD-NARX model is proposed.The input feature set of the corresponding date is deduced on the basis of the existing historical correlation data through the autocorrelation function(ACF),and combined with the time regular characteristics of short-term load fluctuations.After normalizing the input feature set,the Hankel matrix is used to complete the conversion from the univariate input feature sequence to the multidimensional data matrix,and the dynamic modal estimation and eigen decomposition of the multidimensional data matrix that obtained from the previous step are solved by dynamic mode decomposition(DMD).At the same time,the multi-scale dynamics of the underlying power load will be grasped more deeply.Finally,the nonlinear autoregressive with external inputs(NARX)neural netuork model,and the dynamic modal estimates obtained in the previous step are used as the basis for calculating the load distribution in each period of the corresponding forecast date.And then,the final prediction result is derived.The final test data prove that this method can better improve the prediction accuracy of the model.

关 键 词：自相关函数 HANKEL矩阵 动态模态分解 NARX神经网络 短期负荷预测 

分 类 号：Q939.97[生物学—微生物学]