基于混合核KPLS的工业过程质量预测方法

An Industrial Process Quality Prediction Based on Mixed Kernel KPLS

核偏最小二乘(KPLS)能够有效解决数据间的非线性问题并提高质量预测精度,在工业过程监测和质量预测中得到了广泛的应用。良好的KPLS质量预测模型要求核函数同时具备内插和外推能力。然而,传统的单核核函数只能表现出其中一种能力。为了克服这一缺点,本文提出一种混合核KPLS方法用于非线性工业过程质量预测。然后,通过使用遗传算法对混合核函数参数及权重进行优化选取,提高质量预测精度。最后,通过使用田纳西-伊思曼过程的使用实例,说明了该方法的实用有效性。

Kernel Partial Least Squares(KPLS) can effectively solve the nonlinear problem between data and improve the accuracy of quality prediction. It has been widely used in industrial process monitoring and quality prediction. A good KPLS quality prediction model requires the kernel function to have both interpolation and extrapolation capabilities. However, the traditional single-core kernel function can only exhibit one of these capabilities. In order to overcome this shortcoming, this paper proposes a hybrid kernel KPLS method for nonlinear industrial process quality prediction. Then, by using genetic algorithm to optimize the selection of the parameters and weights of the mixed kernel function, the quality prediction accuracy is improved. Finally, an example of the Tennessee-Eastman process is used to illustrate the practical effectiveness of the method.