Support Vector Regression with Miztures of Kernels Incorporating a PSO Algorithm in Weapon System Cost Prediction

摘要：

Considering the shortcomings of conventional cost prediction methods, support vector regression (SVR) was adopted to establish the cost prediction model of weapon system, which can have a good adaptability to the nonlinear modeling and efficiently solve the problems on the determination of network structure and the situation of over-fitting in neural network methods. In SVR modeling, the characteristics of kernels have great impacts on learning and predictive results. Considering the characteristics for fitting and generalization of two kinds of typical kernelsglobal kernel (polynomial kernel) and local kernel (radial basis function kernel), a kind of SVR modeling method based on mixtures of kernels was adopted. On the other hand, because of the increase of the number of model parameters, the particle swarm optimization (PSO) algorithm was used to adaptively evolve SVR to obtain the best prediction performance, in which each particle represented as a real vector corresponds to a set of the candidate parameters of sVR. Experiments in the practical cases demonstrate that the SVR with mixture of kernels has the better prediction performance than with a single kernel and PSO can obtain the optimal results easily.

关键词： Cost prediction Global kernel Local kernel Particle swarm optimization Support vector regression

作者: Tie-jun JIANG Huai-qiang ZHANG

作者单位: Department of Equipment Economy Management,Naval University of Engineering,Wuhan,P.R.China,430033

会议类型: 国际会议

会议名称: The 2nd International Conference on Vale Engineering and Vale Management(2009)(2009年北京价值工程与价值管理国际会议)

会议地点: 北京

会议语种:英文

页码: 153-158

在线出版日期: 2009-10-16（万方平台首次上网日期，不代表论文的发表时间）