基于EGA的RBF-PLS方法为脉冲萃取过程建模

The RBF-PLS Approach Based on Eugenic Evolution Genetic Algorithm and its Application to Pulsed Extraction Modeling

针对径向基网络(RBFN)结构与参数难以确定的问题,在分析径向基函数-偏最小二乘(RBF-PLS)方法的基础上,提出以模型拟合和预报性能为目标,同时优选RBF的宽度参数和PLS成分数,并设计了基于优进策略的遗传算法(EGA)实施优化。EGA增加了模式搜索寻优算子,对交叉算子作了改进,自适应地调整交叉率和变异率,由此形成EGA-RBF-PLS方法,并将它应用于回收ε-己内酰胺的脉冲萃取过程。它工作量小,效果良好,所建模型的拟合和预报性能明显优于近似机理模型和其它RBFN模型,稳健性也更好。

In order to overcome the difficult problem of selecting the architecture and parameters of Radial Basis Function Network (RBFN), the RBF-PLS approach based on eugenic evolution genetic algorithm was proposed. Its abbreviation is EGA-RBF-PLS. In this method, the expression of radial basic function is changed, and the spread parameter of hidden units of RBFN and the number of PLS components extracted are optimized by EGA. The object function of EGA is the performance of fitting and predicting of the model. Here EGA integrates deterministic optimize operator, i.e. pattern search operator, with GA whose crossover probability and mutation probability change adaptively through generations. The approach was successfully applied in modeling the procedure of pulsed extraction process recovering caprolactam from polluted water. The approach's modeling workload is not heavy, and performance of predicting and stability of the model are superior to that of mechanism model or other RBFN models.