基于主成分分析的舰船装备维修费组合预测

针对基于回归的组合预测模型，由于舰船装备维修费预测时可利用的样本小、可用的单项预测方法多，容易导致预测模型的数量比用于组合预测的样本数量多，出现回归系数无法估计的问题。在建立基于回归的舰船装备维修费组合预测模型前，首先对各单项预测方法预测结果进行主成分分析，建立舰船装备维修费实际值在选取主成分上的回归模型，给出基于主成分分析的组合预测模型；然后针对主成分分析中根据主成分的累积贡献率确定主成分数量具有一定的主观性，建议采用AIC确定主成分的数量；最后，采用实例对给出的方法进行分析和验证。结果表明：在舰船装备维修费组合预测中，该方法不仅解决了预测模型多于用来组合预测的样本数量问题，而且还可以解决单项预测方法之间的共线性问题，且其预测性能明显优于常用的组合预测模型。

Combination Forecasting Method of Ship Equipment Maintenance Cost with Principal Component Analysis

Using regression-based combination forecasting method to forecast ship equipment mainte nance cost has fundamental drawback that the number of methods available is bigger than the cost sam ples’,making it insufficient to estimate the regression coefficient.This paper proposed to apply the Princi pal Component Analysis(PCA)on the results which obtained by the individual forecast method prior to building regression-based combination model of a ship equipment maintenance cost forecast,and then the regression model of a realistic ship equipment maintenance cost on selected principal components was es tablished,further,the forecast combination model based on PAC was presented.Meanwhile,in the pro cess of PCA,there was some subjectivity in deciding the number of principal components according as cu mulate contribution rate of principal components,so we took the Akaike’s Information Criterion(AIC)as a substitution.Finally an exampal was used to analyze and validate the performance of the above mentioned method.The results show that the presented method not only can solve the problem that the number of fore casting models is in excess of the sample’s in the regression-based combination forecasting of ship equip ment maintenance cost,but also can solve the collinear between individual forecast method.And the perfor mance of the presented method obviously superior to the usual combined forecasting models.