Application Of Identification Of Chemical Processes Based On Kernel Learning Method

The process simulation,optimization and control of the complex chemical process are the important means to deal with the chemical process.The premise of good control is to make a precise modeling of the chemical process.Chemical process is often nonlinear,complex and influenced by many factors.It is difficult to establish models by process mechanism analysis.Therefore,based on neural network,fuzzy neural network and kernel learning method,computational intelligence has developed rapidly in the field of nonlinear dynamic identification.The kernel learning method based on statistical learning theory is a new machine learning method.Aiming at typical chemical processes,a kernel learning modeling method based on kernel partial least square(KPLS)and kernel recursive least square(KRLS)is proposed.Two kernel learning methods are applied to pH neutralization process and continuous stirred tank reactor(CSTR)process.In order to verify its effectiveness,it is compared with the existing methods under the same conditions.The experimental results show that both KPLS and KRLS have high dynamic modeling accuracy,and the KRLS method has the best modeling effect.The main research is as follows:(1)In the field of pattern recognition and machine learning,kernel learning is a research hot spot.The pH neutralization process is divided into weak acid and strong base process,strong acid and strong base process and two output process with buffering flow.Based on the analysis of the principles of kernel learning and two typical chemical processes,the existing kernel learning methods such as support vector machine and kernel principal component analysis and their algorithm implementation are studied.(2)The basic principle of KPLS method and its learning algorithm are studied.Partial least square(PLS)applies input variables to latent variables,and uses latent information to extract latent features from the covariance information between input and output variables.Based on the kernel function technique,the KPLS method maps the input into the high dimensional feature space,then constructs a linear PLS regression model,and converts the linear inseparable problem to the linear problem by nonlinear mapping.At the same time,KPLS has the advantages of kernel skills and PLS,effectively eliminating the useless interference information,extracting the potential features between variables to the maximum extent,and realizing many functions such as reducing dimension,converting multiple correlations and eliminating interference information.In view of pH neutralization process and CSTR process,a model based on KPLS method is established.Under the same condition,the identification accuracy of KPLS method is higher than that of existing SVM and KPCA-SVM methods.(3)The basic principle of KRLS method is studied.It uses the sparsity criterion of approximate linear dependency(ALD)and combining the advantages of the kernel function technique,limiting the increase of the dimension of the kernel matrix,reducing the computation complexity and storage capacity,and is suitable for large-scale data set training and dynamic time-varying process modeling.Aiming at two chemical processes,a KRLS based identification model is established.Experiments show that KRLS learning speed is stable and fast,and the modeling accuracy is high.