Research And Application On Multi-model Control Method Based On The Neural Network Sets

In Actual control process, for different uncertain controlled objects, traditional adaptive control method could not obtain satisfactory control effect, or even can't control. The multi-model adaptive control method using factor-synthetic strategy is an effective method which is able to deal with the complex nonlinear time-variable uncertain problems. To solve the problems in the traditional adaptive control methods and the multi-model adaptive control method, this thesis unified the neural network technology and the multi-model control policy to propose a Multi-model Control Method Based on the Neural Network Sets to solve the control problems of nonlinear with traits of time-delay and time-varying parameters. The primary contents are as follows:(1)Research the theory of the Multi-model Control Method Based on the Neural Network Sets. Including the theory of Multi-model Control Method, Neural Network and how to unify the neural network technology and the multi-model control policy;(2)Take the time-varying nonlinear three-tank level control system with uncertainty and time-delay as the experiment platform to research the application of the Multi-model Control Method Based on the Neural Network Sets in the process control system. The main ideas are:identifying the non-linear systems with the neural network taking advantage of its great ability of random approach for non-linear function. Generally speaking, the uncertainties of system are bounded, that is, the parameter change space of non-linear object is limited. Suppose that this limited parameter space is divided to N sub-space, and a neural network model is established for each parameter sub-space. These N neural network models constitute a neural network set, which can used to construct a non-linear system controller, named the multi-model control method based on the neural network sets in this thesis. And with this method a new controller which can effectively control a three-tank water system with traits of time-delay, time-varying parameters and non-linear was designed. Moreover, the smooth switch between the sub-models is also a difficult problem which was unable to avoid, this article also proposed a kind of smooth linear weighting transition algorithm. The control algorithm is realized by the STL (sentence table language) of STEP7. The three-tank level control system were designed on this method shows its good robustness for the variation of the characteristics of the system, and demonstrates the efficacy of this method.(3)Taking the Multi-model Control Method Based on the Neural Network Sets Practiced in an actual project cooperated with China Helicopter Research and Design Institute named the wing of helicopter testing systems. There are three control modes named Local Mode, Remote Manual Mode, and Remote Automatic Mode in a subsystem named Hydraulic Pump Control System. And the Remote Automatic Node demands the motor speed increasing curve varying adaptively with the actual load. So different curves were designed for different load, and were fitted with Neural Network. Because the motor current would different when the load was changed but the curve did not, the motor current is taking as the switch basis of multi-curves and switch it directly. In the thesis, the system Structure, the hardware design and the communication networks are analyzed systematically.