基于混合核函数的OLS软测量建模方法研究

The Study on the OLS Soft Sensor Modeling Based on the Mixed Kernels

针对核建模方法中单一核函数不能准确描述数据的分布特性问题,提出一种基于混合核函数的正交最小二乘(OLS)算法并将其用于工业过程软测量建模。采用混合核函数代替基本OLS方法中的单一核函数,利用混合核函数兼具局部和全局核函数的性能,可以提高模型的泛化能力和非线性处理能力。核参数的选择对模型的影响较大,采用粒子群优化算法对核参数进行寻优。在工业聚丙烯熔融指数软测量模型中的应用结果表明,基于混合核函数OLS方法能够比PLS、基本OLS方法更准确地预测熔融指数的变化情况。

With regard to the problem that single kernel can not accurately describe the distribution character of data in the kernel modeling method, an algorithm of orthogonal least square（OLS）based the mixed kernels is proposed and used for soft sensor modeling in industrial process. To take the mixed kernels replace single kernel in OLS and utilize properties of local and global kernel of the mixed kernel can improve their genelization ability and nonlinear treatment capacity. The selection of kernel parameters has great influence for the model, the particle swarm optimization algorithm is used to carry out selection of kernel parameters. The application of industrial polypropylene melt index soft measurement modeling has indicated that OLS based the mixed kernels can more accurately predict change of index than PLS and basic OLS method.