基于核函数及参数优化的KPLS质量预测研究

The optimization of the kind and parameters of kernel function in KPLS for quality prediction

核偏最小二乘(KPLS)在工业过程监测和质量预测中得到了广泛的应用,核函数和核参数的选取对KPLS质量预测结果有重要影响。然而,如何选择核函数类型和核参数一直是该方法应用的瓶颈。针对以上问题,提出一种改进遗传算法的核函数优化方法。该方法将核的种类及核参数作为优化的决策变量,以均方根误差为目标,分别从编码方案、遗传策略、适应度函数优化、交叉和变异算法等方面进行设计,以保证核函数种类的多样性,利用2折交叉验证法对训练结果进行验证。以田纳西-伊斯曼过程(TE)与MATLAB结合进行仿真实验,仿真结果表明,该方法能寻找到最优核函数以及其核参数,具有很好的稳定性和一致性。

Kernel partial least squares(KPLS)has been widely used in industrial process monitoring and quality prediction.The choice of kernel function and kernel parameters has an important impact on the KPLS quality prediction results.However,how to choose the kernel function type and kernel parameters has always been the bottleneck of the application of this method.To solve the above problems,a kernel function optimization method based on improved genetic algorithm is proposed.In this method,the kernel type and kernel parameters are used as the optimal decision variables,and the root mean square error is targeted.It is designed in terms of coding scheme,genetic strategy,fitness function optimization,crossover and mutation algorithms to ensure the variety of kernel functions,and uses the 2-fold cross-validation method to verify the training results.The Tennessee-Eastman Process(TE)is combined with MATLAB for simulation experiments.The simulation results show that the method can find the optimal kernel function and its kernel parameters,and has good stability and consistency.