Research On T-S Fuzzy Modeling Of Complex Thermal System

In the thermal process, the dynamic behavior of plants shows a characteristic of great delay, big inertia, time variance and non-linearity, which makes the modeling very difficult, and the optimal control for thermal processes impossible. While the T-S model, which is proposed by Takagi and Sugeno in 1985, can approach any nonlinear dynamic system accurately, so it becomes more and more widely used in system identification. According to the objective of building accurate nonlinear model for thermal process, some researches on off-line and on-line fuzzy modeling have been done.1 .Research on off-line T-S fuzzy modelingConsidering the traditional fuzzy cluster algorithm needs a large quantity of calculation, prone to dead center, local minimum and center redundancy, in the iterative optimization of cluster center, a new parameters optimization methods based on chaos genetic algorithm is proposed. The order of the model is identified based on the loss function, and the input variables are selected based on the input selection criteria. In this paper, the extended T-S model is adopted, in which adaptive extended gauss membership function is used, and its figure is optimized by chaos genetic algorithm. Finally the recursive least square method is adopted to identify the conclusion parameters of the fuzzy model. The chaos genetic algorithm introduces chaotic immigration operator into genetic algorithm. The immigrations will replace the bad individuals of the original group, and involve in the mating, in order to ensure the diversity of the group and prevent the gene recession. This algorithm can overcome the shortcoming of early mature and slow convergence. And it shows great performance in antecedent parameters estimation.2. Research on on-line T-S fuzzy modelingConsidering the uncertainty of the operating conditions of actual thermal processes, the changes of external environment and the drift of parameters and structure, off-line modeling is difficult to reflect the non-linear system adaptively. On-line fuzzy modeling, therefore, has a strong theoretical and practical value, and it is widely used in adaptive control and predictive control. A new way of on-line identification based on an improved T-S model is presented. The clustering centre vectors are updated by the close degree, which indicates the relation between input vectors and clustering centre. The input data space is partitioned into some local regions by the distance between input data and clustering centre. The conclusion parameters are identified by the recursive least-square identification algorithm. This identification algorithm can build the .T-S model with few rules and high-precision, and it is simple and easy to carry out.Finally, the effectiveness and practicability of these methods are demonstrated by the simulation results of the Box-Jenkins model and the overheated stream system.