Multiple Model Soft Sensor Based On Improved Affinity Propagation Clustering

In modern complicated industrial process,some variables are difficult to be measured or even can not be measured on line by traditional instruments and sensors.When soft sensor is used for unpredictable variables,single model can't effectively describe the characteristics of working conditions normally.Multiple models can describe complex production process better.This paper classifies the training samples into several classes using clustering algorithm,and train the sub-models according to corresponding sub-class samples.The test samples are assigned to appropriate sub-class.The degrees of membership is used for combining several models to obtain the finial.Clustering algorithm determines the accuracy of modeling directly,and modeling algorithms also have a significant impact on the modeling results,and outliers have a great impact on soft sensor modeling algorithm that needs to be processed.This paper do some work from the data clustering and sub-model modeling and outlier processing method.Main works are as following:In this paper,we improve the clustering algorithm,sub-modeling methods and outliers processing methods:Firstly,the modeling effect that based on data-driven modeling approach depends on the assumption that conform the data characteristics.In order to study on the scope of application,we investigate frequently-used soft sensor methods based on several sets of data and get some useful conclusions.Secondly,the wavelet kernel function can approximate more complex nonlinear function and improve parameter estimation;orthogonal least squares algorithm can extract the nonlinear relationship between the main component and output variables.Construct orthogonal least squares based on wavelet kernel function method can improve the generalization ability of the model effectively.Particle Swarm Optimization is used to optimize kernel parameter.Thirdly,to the problem of outliers existing in the operating process,this paper introduces the method of outlier detection based on OCSVM.To determine whether the outliers we get is authentic,the concept of outliers classification is put forward.Based on the Bayesian classification principle,varieties of outliers include the impulse outliers,the short-step type outliers and the step outliers.Fourthly,A multi-model soft sensor modeling method based on an improved adaptive affinity propagation clustering is proposed.The method first finds out the range of preference,then searches the space of preference to find a good value which can optimize the clustering result.The degrees of membership is used for combining several models to obtain the finial.Finally,the final part of this paper summarizes the research working on soft measurement technology of multiple models and puts forward some questions in need of further research.