| In the industrial process,due to conditions such as on-site environment and detection technology,some important variables cannot be directly measured by online instruments,and traditional laboratory offline analysis has serious time lag and other problems,which cannot meet the real-time requirements of industrial control.Soft-sensing technology is to realize the real-time estimation of key variables by constructing mathematical models between measurable variables and key variables.For industrial processes with strong coupling and multi-mode characteristics,this paper is based on the Gaussian Process Regression(GPR)algorithm and studise and improvse the soft-sensing modeling method from the aspects of auxiliary variable selection and multi-model modeling.The main contents of this paper are as follows:(1)Aiming at the problems of coupling,correlation and redundancy between various variables in the actual industrial process,a GPR modeling method based on variable selection and combined covariance is proposed.Firstly,the random forest(RF)algorithm is used to give the importance score of the auxiliary variables,and the variables with high importance scores are selected as the input of the model to reduce the correlation and redundant information between the auxiliary variables;Afterwards,Gaussian process regression model are built using the combined covariance function instead of the single covariance function to better capture the various linear and nonlinear features in the data.Finally,the simulation results of the sewage treatment process data show that the proposed method has a good predictive effect.(2)Considering that it is difficult for the single soft-sensoring model to describe the multi-mode characteristics of industrial processes,a multi-model soft-sensoring modeling method based on improved density peaks clustering(DPC)is proposed.Among them,the K-nearest neighbor algorithm is applied to calculate the local density to avoid the shortage of artificial selection of the cutoff distance and the influence of the local density calculation method by the size of the data set.The K-nearest neighbor algorithm and the weighted K-nearest neighbor algorithm are used to improve the remaining point allocation strategy to avoid chain allocation errors;At the same time,the posterior probabilities of the new sample belonging to each sub-model are calculated adaptively based on the prediction performance of the sub-model and combined with Just-in-Time learning,and which are used to merge the prediction values of each GPR sub-model to obtain the final output.The proposed model is validated to achieve higher prediction accuracy by simulating the standard dataset and the sulfur recovery unit data.(3)In order to solve the problem of difficult measurement of key parameters in the wastewater treatment process,a soft measurement platform for sewage treatment is built based on virtual instrument software Lab VIEW and Python together.Firstly,Python is applied to establish several soft-sensing models with different algorithms based on the sewage treatment process dataset;After that,Lab VIEW is used to collect the corresponding auxiliary variable data as the input of the models;Finally,different algorithmic models are invoked in the soft measurement platform to achieve online measurement and offline analysis for key parameter of effluent BOD content,and improves the real-time and accuracy of industrial monitoring. |
