A hybrid algorithm and its applications to fuzzy logic modeling of nonlinear systems.
详细信息
文摘
System models allow us to simulate and analyze system dynamics efficiently. Most importantly, system models allow us to make prediction about system behaviors and to perform system parametric variation analysis without having to build the actual systems. The fuzzy logic modeling technique has been successfully applied in complex nonlinear system modeling such as unsteady aerodynamics modeling etc. recently. However, the current forward search algorithm to identify fuzzy logic model structures is very time-consuming. It is not unusual to spend several days or even a few weeks in computer CPU time to obtain better nonlinear system model structures by this forward search. Moreover, how to speed up the fuzzy logic model parameter identification process is also challenging when the number of influencing variables of nonlinear systems is large. To solve these problems, a hybrid algorithm for the nonlinear system modeling is proposed, formalized, implemented, and evaluated in this dissertation. By combining the fuzzy logic modeling technique with genetic algorithms, the developed hybrid algorithm is applied to both fuzzy logic model structure identification and model parameter identification. In the model structure identification process, the hybrid algorithm has the ability to find feasible structures more efficiently and effectively than the forward search. In the model parameter identification process (by using Newton gradient descent algorithm), the proposed hybrid algorithm incorporates genetic search algorithm to dynamically select convergence factors. It has the advantages of quick search yet maintains the monotonically convergent properties of the Newton gradient descent algorithm. To evaluate the properties of the developed hybrid algorithm, a nonlinear, unsteady aerodynamic normal force model with a complex system involving fourteen influencing variables is established from flight data. The results show that this hybrid algorithm can identify the aerodynamic model structures much quicker than the forward search. In addition, the results also show that this hybrid algorithm can identify model parameters much quicker than the one with fixed and arbitrary convergence factors. Finally, an application of the fuzzy logic modeling technique to Kansas Arbuckle oil well performance analysis is performed. It gives oil operators a powerful decision-making tool for candidate-well selection and treatment to optimize performance.