Intelligent Soft Sensing Based On Nonlinear Feature Extraction

With the development of modern industrial process requirements of control, metering,energy efficiency and operation reliability, the demanding of various quality variablemeasurement is increasing. In modern complicated industrial process, some variables are veryhard to be measured or even cannot be measured on-line by existing instruments and sensors. Atpresent, soft sensing has become one of the most important research areas in process controlfield. This paper proposed several soft sensing methods from the studies of nonlinear featureextraction and intelligent soft sensing. In order to reduce the complexity of the soft sensing, weadopt nonlinear feature extraction algorithm. Then the intelligent soft sensing is aopt to improvethe model accuracy. Furthermore, the problems of soft sensing implementing for FCCU mainfractionator are also studied. The detailed content is arranged as follows:A novel nonlinear feature extraction algorithm of KIsomap is proposed for the nonlinearcharacteristics and high dimensionality. On the one hand, KIsomap algorithm preserves themanifold structure of data space. On the other hand, KIsomap algorithm can projected the testdata into low dimensional space. Otherwise, considering the existence of non-stationary andfuzzy information in input data, a novel feature extraction based on fuzzy informationgranulation is proposed. By dividing the window and fuzzy, massive data is processed usingtriangular fuzzy particle. Then the data of nonlinear feature extraction which are Low、R and Upare gained. Simulation results show that this model can get a better prediction. These propsedmodeling are used in nonlinear feature extraction of diesel oil solidifying point. The resultsshow that these models are effective data modeling.For intelligent soft sensing aspect: to accelerate the learning speed and reduce the presetedparameters of soft sensing, an intelligent soft sensing based on generalized dynamic fuzzyneural network is proposed. The parameter adjustment and structure identification can proceed in parallel. Furthermore, its learning speed is quite fast and its preseted parameters are less thanfuzzy neural network. The proposed method is used to build soft sensing of pH neutralizationprocess. The result testifies that this method is very simple and practical. Otherwise in order toovercome the difficult problem of determining the optimum parameters of least square supportvector machine (LSSVM), a kind of intelligent soft sensing based on improved GA-LSSVM isproposed. This model used the powerful global search performance of adaptive GA. Thereforethe adaptability of LSSVM model is advanced. The result shows that IGA-LSSVM approachhas high precision and good generalization ability.Lastly the application results on FCCU main fractionator are introduced. A novelintelligent soft sensing based on fuzzy information granulation and improved GA-LSSVM isproposed. The fuzzy information granulation has excellent performance of nonlinear featureextraction which are Low、R and Up are gained. Then improved GA-LSSVM is applied toproceed regression modelling. The result shows a better prediction accuracy and generalizationcapability. Moreover this modeling can be used to guide production efficiently.