Study On Adaptive Control Based On Fuzzy-Neural Networks

Fuzzy-neural network (FNN) is composed of the neural network(NN) and fuzzy logic system with the function of good at self-learning and self-tuning. The theory and application of FNN are developed very quickly, and i s an active branch in the intel1igent control, but there are still some problems existed. In this dissertation, the following problems of FNN are researched.1. The traditional control system based on FNN, either does not introduce systematic feedback information to the FNN controller(FNNC), which may lead to the slowed convergence rate of the control system because of lacking transient information of the system, or introduces systematic feedback information to the FNNC directly which makes the number of the inputs to FNNC increased, and the complexity of the network is increased. This may reduce the convergence rate of the FNN . In order to solve this drawback, a control system model appended a network to gathering the output information is proposed Simulation results show that the systematic convergence rate improved objectively.2. The problems of FNNC as membership functions and fuzzy rules in fuzzy reasoning can be sum up as the structure of NN selecting and the parameters of NN training. Usually, the fuzzy space is divided subjectively. The size of the fuzzy space is very difficult to confirm if lacking of experience or expertise of the system. This may cause some important input signals goingbeyond the fuzzy space. Also the number of the fuzzy partition is very difficult to fix . On one hand , if grade is too large , the control sensitivity will be reduced , and it' s very difficult to reach the required precision, on the other hand, if grade is too small, the calculating complexity of the NN increased greatly. This is unfavorable to convergence of NN , and even cause unstable. In order to solve these two problems , a structure modified FNN is proposed. Adopt a small-scale FNN initially , and with the process of controlling, the structure of the FNN (nodes of the second layer and the third layer) is modified adaptively. This will improve the performance of FNNC.3. The large account of computing work of updating weights and long training time usually limit the application of FNNC in industry. Moreover, when it is trained on-line to adapt to plant variations, the over-tuned may cause system oscillate extensively. A kind of optimization method is proposed. The updating step is adjusted adaptively in accordance with the error and the change of error of the system based on the T~S model to get better performance. Simulation results show that the training time is reduced greatly , the system' s robustness is strengthened and theconvergence rate is speed up.