基于加权偏最小二乘回归的中长期负荷预测简

针对中长期负荷预测，考虑各历史样本在建立适用于预测对象的模型时处于不同的地位，应分配不同的权值，提出一种基于加权偏最小二乘回归（weighted partial least squares regression，WPLSR）的预测方法。利用相似离度计算历史样本与预测对象的相似度，判定样本是否含有异常值，自适应地为历史样本分配权值，进而采用偏最小二乘回归（partial least squares regression，PLSR）提取主成分和回归分析。算例结果表明WPLSR方法的预测精度比普通PLSR模型有显著提高，具有良好的可行性和有效性。

Mid-long term load forecasting based on weighted partial least squares regression

Considering that historical load samples each have asymmetrical status and should be assigned to different weightings in the med-long-term load forecasting, this paper pro-poses a weighted partial least squares regression（WPLSR）algo-rithm. The specific modeling procedures are：Analog Deviation be-tween the historical samples and predicting samples is computed;identify abnormal samples; adjust sample weights; partial least squares regression analysis. Experimental results show that the pre-diction accuracy of the weighted partial least squares regression （WPLSR）algorithm is remarkably higher than that of traditional PLSR model. the proposed modeling method is practicable and ef-fective.