This paper deals with identification and control of a highly nonlinear real world application. The performance and applicability of the proposed methods are demonstrated for an industrial heat exchanger. The main difficulties for identification and control of this plant arise from the strongly nonlinear center and the widely varying dead times introduced by different water flows. The identification of this three input one output process is based on the local linear model trees (LOLIMOT) algorithm. It combines efficient local linear least-squares techniques for parameter estimation of the local linear models with a tree construction algorithm that determines the structure of their validity functions. Furthermore, a subset selection technique based on the orthogonal least-squares (OLS) algorithm is applied for an automatic determination of the model orders and dead times. This strategy allows to design a wide range high accuracy nonlinear dynamic model of the heat exchanger on which the predictive control approach is based on. The nonlinear predictive control takes the speed and limit constraints of the actuator into account and leads to a high performance control over all ranges of operation.
The paper represents a new aproach to the predictive model-reference control. The prediction of the process output signal is made on the basis of fuzzy process model. Using the fuzzy model of the process the forecast of the process output over a certain horizont in the future is cMculated and can be used as a predictor in the long-range predictive control strategy. The concept is implemented on real industrial scale temperature plant.
KEYWORDSpredictive control, fuzzy relational model, identification, nonlinear systems
1.INTRODUCTIONThe predictive control has become a very important area of research in recent years. The principal is based on the forecast of the output signal y at each sampling instant. The forecast is made implicitly or explicitly according to the model of the process to be controlled. In the next step the control is selected which brings the predicted process output signal back to the reference signal in way to minimize the area between the reference and the output signal. The fundamental methods which are essentially based on the principal of predictive control are Richalet's method (Richalet et al., 1976, Model Algorithmic Control), Cutler's method (Dynamic Matrix Control), De Keyser's method (Extended Prediction Self-Adaptive Control) and Ydstie's method (Extended Horizon Adaptive Control).According to the process model two main approaches have been developed in the area of predictive control. The first one is based on parametric model of the controlled process. The parametric model could be described in form of transfer-function or in state-space domain. An important disadvantage of using the parametric model is that it represents a linearized model of the process. The control of the strong nonlinear processes could be unsatisfactorilly. The second approach proposed in literature is based on nonparametric model. The advantage of this approach is that the model coefficient can be obtained directly from samples of the input and output responses without assuming the model structure. In our example a scheme of predictive control based on fuzzy relational matrix model is proposed, which represents a combination of nonparametric and parametric approach to the predictive control.Predictive control based on fuzzy relational matrix model is capable to control also very difficult processes, such as nonlinear processes, processes with long time delay and non-minimum phase. The controllers based on prediction strategy Mso exhibit remarkable robustness with respect to model mismatch and unmodeled dynamics.The first part of the paper deals with the concept of fuzzy relational matrix modelling. In the second part the concept of fuzzy predictive control is given. Finally, the implementation of fuzzy predictive control on the real temperature plant is presented.$931
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.