Multi-model Soft-sensor Based On Fuzzy Clustering And Its Application

Due to the complexity of industrial processes and the current technology development level,many important process variables are difficult to achieve real-time on-line measurement.In order to solve such problems,soft-sensing technology gradually developed.The basic idea is to estimate the process variables that are difficult to measure directly based on the easily measurable process-dependent variables which are associated with the variables to be measured.In this paper,two modeling methods based on fuzzy clustering are proposed,which are verified by MG chaotic time series and Box-Jenkins boiler data respectively.Finally,the model is applied to the SCR inlet NOx Estimated and has achieved good results.The main work of the paper has the following aspects.(1)Firstly,the background of soft-sensing technology and its research status are introduced.Secondly,the basic idea and basic steps of soft-sensing modeling technology are analyzed,and several modeling methods commonly used in soft-sensing modeling are introduced briefly.(2)Several modeling methods commonly used in soft-sensing modeling are introduced in detail,such as PCA,PLS,ANN and SVM.Based on the introduction of its basic principle,the advantages and disadvantages are analyzed.(3)Introduced the fuzzy theory and several fuzzy clustering algorithms,and proposed a soft-sensing method based on adaptive fuzzy clustering.Based on the TS model structure,the GK clustering algorithm is first designed into a recursive form,and then adaptive clustering is implemented according to certain criteria so as to realize parameter identification of the antecedent of the model.Then the RLS is used to identify the parameters of the sub-models,and finally the output of each sub-model integrated to get the final result.The model is validated by Box-Jenkins boiler data and compared with the results of the literature.(4)The basic idea of simplified T-S model is introduced,and a soft-sensing model based on simplified T-S model is established.The simplified T-S model avoids the complexity of traditional T-S model identification,further simplifying the model and achieving an acceptable range of accuracy.In this paper,the input data is clustered on-line in the recursive form of the subtractive clustering algorithm and the clustering center parameters are updated in real time to obtain the antecedent parameters of the model.Then the global RLS algorithm is used to identify the sub-model parameters,and finally the output of each sub-model is integrated to get the final integrated output.Taking the chaos time series of MG as a verification example,the validity of the model is verified,and it is applied to NOx estimation of SCR entrance of power station boiler.The good effect is achieved.