KPCA-模糊加权LSSVM预测方法及其应用

摘要

Aimed at the problems existed in the industrial systems such as slow variation in parameters, noise in field data, and multiple correlation between variables, an improved LSSVM prediction algorithm based on fuzzy weighted and Kernel Principle Component Analysis (KPCA) is proposed. In the algorithm, KPCA is used to eliminate the noisy of data, lower the multiple correlation and reduce the dimensions of the input samples. And to demostrate the contribution of the parameters for the system, a fuzzy weighted method is introduced in the algorithm. The parameters of LSSVM model is optimized by improved particle swarm algorithm. The experimental verification analysis is performed using the industrial production data from purification process. The simulation result shows that the proposed algorithm satisfies the requirements of cobalt ion concentration prediction in industrial field, and has a higher accuracy compared with Principle Component Regression ( PCR) or LSSVM.%针对工业系统参数缓慢变化和工业现场数据存在噪声和变量间存在多重相关性,提出了一种KPCA-模糊加权LSSVM预测方法;采用KPCA去除数据的噪声和降低样本数据各参数间的多重相关性,减少样本数据维数;针对不同时期样本数据重要程度不一样,提出了模糊加权的思想;利用改进微粒群算法(PSO)优化LSSVM模型的惩罚因子和核函数参数;以净化除钴过程生产数据进行验证分析,仿真结果表明,KPCA一模糊加权LSSVM预测模型精度高于主成分回归(PCR)和LSSVM,能满足工业现场钴离子浓度预测的要求.