System Identification For Industrial Control And Soft Sensor Modeling

Model building based on observable information is an effective way to further know and study a system.In the fields of industry and aerospace,models are used to design appropriate control methods,predict the output and optimize objectives for the system.System identification is an effective way to obtain dynamic models of sys?tems and research on system identification has great value in theory and applications.In this work,firstly,closed-loop identification for two most common industrial con-trol methods,PID and MPC closed-loop system,are studied.A method of closed-loop identification without using external excitation is proposed from the perspective of re-ducing the cost of identification experiments,where theoretical informative conditions for closed-loop identification without using test signals are derived.Then,this paper expands the application of system identification in the area of soft sensor modelling.The main research contents and results are summarized as follows:1.Closed-loop identification based PID controller tuning is studied.A method of closed-loop PID tuning without using test signals is developed.Two important concepts in closed-loop identification,data informativity and model identifiabili-ty,are introduced.Through the two concepts,the question whether it is possible to estimate the model of system from the data under the situation where there is no external excitation is explored,and it is shown that the model can be esti-mated only if the controller degree is higher than that of the model.For the PID controlled close loop,the controller degree is low,hence the persistent excitation condition is not always satisfied.To solve this problem,a method of identification test based on switching controllers is proposed.This method can not only make the data informative enough to ensure the model is identifiable,but also improve the control performance and reduce the identification cost during the identifica-tion experiment stage.Then,a complete identification based PID tuning strategy without using external excitation is proposed.The effectiveness of the proposed approach is illustrated by simulations.2.Sufficient informative conditions for a data set in multi-variable closed-loop sys-tems are obtained for the identification of commonly used model structures.It is concluded in existing literatures that if the controllers are of sufficient complexi-ty then the data set generated from the closed-loop system would be informative,which is qualitative and not useful.In this work,using the number of inputs and output.s and order of model and controller to quantify the model and controller complexity,the quantitative informative results are derived under both the situ-ation where the closed-loop system is controlled by a linear time-invariant con-troller and by switching controllers.These results provide a theoretical basis for studying the identification of multi-variable systems without using test signals.3.Multi-variable MPC closed-loop system is studied.A method of identification test based on switching MPC controllers is proposed to reduce the cost of identi-fication experiment.The equivalent minimum form of MPC is derived when there is no constraints or the constraints are inactive in the objective function of MPC,and through this minimum form,it is shown that the informative condition would not always be satisfied.Then,the relation between the control performance and the weighting matrix when there is a model mismatch in MPC closed-loop system is discussed.Based on this relation,the method of identification test using switch-ing MPC controllers is proposed.Similarly,through this switching controllers based identification experiment,the multi-variable model can be identified and the control performance can be improved simultaneously.The effectiveness of the proposed method is veri:fied by simulation experiments.4.An identification algorithm for soft sensor modeling using a linear parameter varying(LPV)model with multiple model structures is presented.Firstly,the output error(OE)method is used to estimate model parameters,and a numeri-cal optimization algorithm based on relaxation strategy is developed,which can guarantee the numerical stability.Then,for LPV model structure determination,an engineering approach is proposed,which combines process knowledge with the so-called final output error criteria(FOE).The method is verified using both simulation data and industrial data.In the industrial case study,the LPV models give more accurate prediction of product qualities than that of a linear dynamic model and that of a static nonlinear model.