Soft-Sensor Modeling For Freeness Estimation Of High-Consistency Refining System Based On Improved Support Vector Regression

As a key step of chemi-mechanical pulping,high-consistency refining pulp is directly related to the quality and energy consumption in producing process,which belongs to the canonical process with high energy consumption and low efficiency.For this purpose,the research on the modeling and control of high concentration pulping process is of great significance for the control quality and energy saving of the pulping process.The mechanism of high concentration pulping is so complicated that it is difficult to establish a precise mathematical model,and it has many features such as multi variable,strong coupling,large time delay and nonlinearity.As one of the key indexes for the evaluation of pulp quality,freeness is closely related to the quality of the pulp and the energy consumption of the whole process.However,freeness is difficult to use the existing measurement methods for direct online detection,at present only manual off-line test method can be used,but the period is too long that it is difficult to meet the real-time monitoring and real-time control requirement of high-consistency refiner freeness.Therefore,effective measures must be taken;the soft sensor model of the high consistency pulp system was established according to the key production process of measurable parameters,achieve the online continuous estimation of the freeness,and it provides technical support to improve the technology and equipment of the high consistency pulping further,and energy saving of the papermaking process.For above problem of modeling soft sensor for freeness about high-consistency refining system of chemi-mechanical pulping,relying on the national natural science fund key project"Towards minimized costs and fiber state distribution optimal operational control for pulping processes",this thesis using data-driven modeling to study the soft sensor for freeness of high-consistency refining system based on improved support vector regression.The main work of this thesis is as follows:(1)A soft sensor modeling for freeness of high-consistency refining system based on improved s-SVR.Firstly,the collected data is processed by data compression,outliers removal and data standardization.Then principal component analysis is used to extract the feature of the data for the modeling sample set.Gauss radial basis kernel function with superior local performance and polynomial kernel function with superior global performance are used to construct the multi-kernels function,so MK-s-SVR method is proposed.Secondly,due to the choice of regularization parameters and kernel parameters for the learning ability and generalization ability of model has a critical influence,the method based on cross validation to select model structure parameters set of MK-?-SVR(CV-MK-?-SVR)is proposed,which can improve the learning and generalization ability of the model.For cross validation search and calculation of time-consuming and accuracy in parameter is not high,such as increasing the precision of parameter,and optimization time increases into exponential stage,so the method based on the adaptive mutation particle swarm optimization algorithm to select MK-?-SVR model structure parameters set(AMPSO-MK-?-SVR).When simulating,use UCI data set of daily bike sharing counts about Capital Bikeshare system in Washington D.C.to verify the effectiveness of the proposed method firstly.And the results show that AMPSO-MK-?-SVR has good advantages of optimizing efficiency and accuracy of the model than CV-MK-s-SVR.Then,the methods are applied to the soft sensor modeling for freeness of high-consistency refining system in chemi-mechanical pulping plant.The results show that the AMPSO-MK-?-SVR has good advantages in searching efficiency and model accuracy.(2)A soft sensor modeling for freeness of high-consistency refining system based on improved LS-SVR.Because the solving process of ?-SVR is inefficient and the accuracy of regression remains to be further improved,LS-SVR is used to improve the solution efficiency and the solution accuracy of the model.So modeling method of LS-SVR-ARMA2K is proposed.Firstly,build auto-regressive and moving average model is built to describe the dynamic characteristics of the research object;secondly,use the multi-kernels function to improve the LS-SVR for improving the solving efficiency,accuracy and the scope of application;then,use cross validation to select the structure parameters set of the model,CV-LS-SVR-ARMA2K proposed;finally,use the adaptive particle swarm optimization algorithm to select structure parameters set of soft sensor model,AMPSO-LS-SVR-ARMA2K proposed;.When simulating,also use UCI data set to verify the effectiveness of the proposed method.The results show that AMPSO-LS-SVR-ARMA2K has great advantages in model accuracy and computational efficiency.Then,the methods are applied to the soft sensor modeling for freeness of high-consistency refining system in chemi-mechanical pulping plant.The results show that LS-SVR-ARMA2K has the obvious advantages in the model accuracy and solving speed compared with the MK-?-SVR and BP-ANN.Beyond that,LS-SVR-ARMA2K has a higher accuracy in the small sample case.