一种改进的CPSO-LSSVM软测量模型及其应用

Soft sensor modeling method based on improved CPSO-LSSVM and its applications

针对最小二乘支持向量机(LS-SVM)在处理大规模数据集的回归和分类问题时缺少支持向量所具有的稀疏性和难以确定最佳模型参数值的问题,提出一种改进算法,利用样本间马氏距离分析样本相似程度,剔除部分相关样本,对样本集进行约简,以恢复LS-SVM的稀疏性,进而利用具有较强全局搜索能力的混沌粒子群优化算法(CPSO)对LS-SVM建模过程中的模型参数进行优化选择,以提高模型的拟合精度和泛化能力。将提出的改进算法用于湿法脱硫系统浆液pH值的软测量建模,给出了应用该方法的具体步骤,研究结果表明,该算法取得了较高的建模精度和泛化能力,为pH值的在线实时监测提供了一个有效手段。

Aiming at the fact that in dealing with the regression and classification problems of large scale data set with a lot of samples, traditional least squares support vector machine （LS-SVM） algorithm has some demerits, such as, the loss of sparseness and the difficulty in selecting model parameters, an improved algorithm is proposed to overcome the above drawbacks in this paper. Mahalanobis distance among samples is used to analyze the sample similarity, eliminate some related samples, perform reduction on the samples and recover the sparseness of the least squares vec- tor machine. Moreover, the chaos particle swarm optimization （CPSO） that has strong global searching capability is introduced to optimize the model parameters in the LS-SVM modeling process to improve the fitting accuracy and en- hance its generalization ability. The improved algorithm was used to establish the soft sensor model for the slurry pH value of WFGb. The detailed procedures of applying this method are given. The research results show that the im- proved algorithm achieves high modeling accuracy and good generalization ability, which meets the industrial meas- urement requirement and provides an effective means for the online and real time pH value monitoring.