jiang hainu jiang

Feature extraction is an important task before weapon system cost forecasting modeling, which affects the forecasting performance of the model. In this paper, feature extraction in the weapon system cost forecasting was studied. In regard to the mechanism of feature extraction and the good performance of support vector machine (SVM), principal components analysis (PCA) and kernel principal components analysis (KPCA) were compared and the SVM-based cost forecasting model was adopted. A cost forecasting modeling framework based on KPCA&SVM was established. At the same time, three cases of cost forecasting, SVM, PCA+SVM and KPCA + SVM, were compared. In addition, considering the consistency of feature extraction and the establishment of cost forecasting model, a joint optimization method based on particle swarm optimization (PSO) was adopted, which can simultaneously achieve feature extraction and the optimization of cost forecasting model. And the characteristics and advantages of the kernel method were analyzed. The calculation results show the good application effect and prospect of feature extraction based on KPCA in the weapon system cost forecasting.