基于偏最小二乘回归的年用电量预测研究

STUDY ON THE LONG TERM PREDICTION OF ANNUAL ELECTRICITY CONSUMPTION USING PARTIAL LEAST SQUARE REGRESSIVE MODEL

对年用电量的预测若采用一般最小二乘回归法建模,其估计参数存在着很大的误差且物理意义明显不足。而偏最小二乘回归方法则实现了多元线性回归、主成分分析和典型相关分析的综合、克服了自变量之间的多重相关性的问题,因而更具有先进性,其计算结果更为可靠,在实际系统中的可解释性也更强,且方法简单,计算快捷。该文将偏最小二乘回归模型(Partial Least Square Regression,PLS)应用于年用电量预测,并与基于最小二乘的多元线性回归模型预测成果进行对比,探讨了偏最小二乘法在电力负荷预测中的可行性和优势。通过四川省电网年用电量预测表明:偏最小二乘回归法比一般最小二乘法优,具有较强的实用性。

The method frequently used in prediction ofannual electricity consumption is least square method (LSM). If there are multiple correlation factors in the multiple linearregressive equations (MLRE), the estimated regressiveparameters with lsm will induce a good deal of errors and theregressive equation reflects no more physical meaning. Thepartial least square method (PLS), proposed in this paper, is acomposition of regressive analysis, main components analysisand typical correlation analysis. This method can easily solve the multiple correlation problems in MLRE analysis with fastcalculation. The estimated regressive parameters with PLS arerobust. A long term prediction of sichuan province annualelectricity consumption, as a case study, has been done. Theresults show that the accuracy is higher than those based on LSM. More advantages in using PLS observed during the study overLSM.