Model Predictive Control And Its Applications In Distributed Parameter Systems

Model predictive control has the capability of dealing with systemconstraints explicitly, as well as easiness of implantation. Currently,application ifelds of predictive control have extended from linear systems to allkinds of nonlinear systems. For the prevalent distributed parameter systems,often characterized by infinite-dimension, they are described by partialdifferential equation model which do not support direct usage of predictivecontrol. So far, development of predictive control for distributed parametersystems is far behind predictive control for lumped parameter system. In orderto solve the problem of flow and temperature field control in practicalengineering, closely connected with the actual object of industrial process,different modeling methods are adopted and predictive control is applied todistributed parameter systems. The main contents of the dissertation are asfollows:First, a multi-scale modeling method for distributed parameter systemsbased on wavelet collocation is given. Quasi-Shannon wavelet is selected andused as primary function in space so that distributed parameter systems areactually converted to lumped parameter systems on collocation point. ForwardEuler Method was used to discretize time variable to obtain the auto regressionmodel of the system under investigation, which guaranteed the sparsity ofprimary function of wavelet, and avoid the defects of computation complexitycaused by lacking clear expression of traditional Daubechies wavelet.Symmetric extension method was adopted to restrain the boundary effect whendealing with system boundary problems, then, the method above was appliedto nonlinear distributed parameter systems described by Burgers equation,comparative results were given on different scales to illustrate the effectivenessof the proposed method.Then, predictive control methods for two classes of nonlinear distributedparameter systems based on model of wavelet collocation were proposed. First,for a class of nonlinear first order hyperbolic systems, first order spatial partialderivative was projected on primary function of quasi-Shannon wavelet, andpredictive model of the system under investigation was obtained through agroup of ordinary differential equations obtained by discretizing time variable.Through the method above, the nonlinear predictive controller was designed,and applied to the heating system of long tube based on flow control, in thisway, the problem of high-order computation model due to lack of dominantpole for hyperbolic type system was overcome. Second, for a class of nonlinear parabolic type system, first order and second order spatial partial derivativewere projected to quasi-Shannon wavelet respectively, eliminating the need ofknowledge of solution of dominant pole of the system. In this way, thecorrespondent lower order model was obtained. A group of ordinary differentialequations obtained through discretizing time variable was selected as thepredictive model of the system and the corresponding nonlinear predictivecontroller was designed. This method is applied to the transfer-reaction systemof catalytic rod putted in the reactor, simulation results indicates that theproposed method can be satisfied with the requirements of system control, andthe model is simple, easy to achievement.And then, for a class of nonlinear parabolic systems, one type of predictivecontrol algorithm based on time and space separation was investigated. First,according to theory of time and space separation, the leading primary functionin space was obtained by Karhunen-Loeve decoupling, then method of leastsquare support vector machine was used to identify and obtain the systemprediction model. Second, standard quadratic optimization performance indexis selected, the system is transformed into an optimization problem withconstraints scrolling, and the nonlinear predictive controller is designed basedon lower order model, the input-output constraint of system is processedeffectively, the proposed method is applied into control of tubular reactor indiffusion-reaction process, and the effectiveness of the method is verifiedcompare with existing algorithm.Finally, the problem of controlling simulation system of temperature fielddistribution in the environment of underground extra-deep well wasinvestigated. Analysis and control of the characteristic of temperaturedistribution of such system has always been a challenging task in the field ofengineering since it has large draw ratio, large time-delay and very complexheat transfer mode at the same time. After an in-depth analysis of themechanism of heat-transfer and numerical studies, the distributed parametermodel of temperature field was given. Subsequently, the low-dimensionalprediction model was obtained by space-time decomposition method, theactual control parameters were selected to design the predictive controller fortemperature field. Finally, The Controller was realized on the Compactlogixsystem, and the problem of high precision control was solved for temperaturefield of the simulation system through the actual operating results.