The Research Of Neural Networks Control Algorithms Without Identification Of Coupling Function

Only by knowing coupling functions of MIMO systems, can control algorithms be used in engineering practice. Therefore it is necessary for most MIMO control algorithms to identify the coupling functions. But the identifications of the coupling functions are toilsome and time-consuming. For some systems, the identifications are very difficult. The thesis establishes a MIMO control algorithm with neural network coupling compensators. The algorithm needs not identification of coupling functions, thus the difficult problems can be avoided.The thesis relates a control algorithm based on parallel coupling compensators. The algorithm produces the control system in which every control channel has a master controller and a parallel compensator. Since the algorithm needs not knowing the coupling functions of the system, it needs not identification of the coupling functions in applications.However, the compensator should use same grade energy as the plant in the system, so that the algorithm is not practical.My thesis significantly improves the algorithm based on parallel coupling compensators. I make the output of coupling compensator to operate actuator in the control system through transforming block diagram of the system, so that the compensator needs a little energy. Every coupling compensator is made up of a neural network and a virtual plant in series. The compensator has no anticipatory element and pure differential element, thus it can be easily realized by using electronic devices.I deduced the formulas of the algorithm to prove its correctness by analytical mathematical method. Because I cannot find any literature about my algorithm, I independently deduce the algorithm under the guidance of my supervisor. The algorithm, I think, is innovative.In a control system based on my algorithm, master controller can be various available controllers, such as PID controller; coupling compensator can be various available neural networks. The thesis select RBF, BP and CMAC as coupling compensators. I have done simulation researches on the systems based on the neural networks. Simulation results testify that my algorithm can be implemented and can improve performance of MIMO control systems, especially, the master controllers and coupling compensators can be designed and made even if the coupling functions are unknown.The thesis provides an advanced control algorithm for process control of chemical industry, petroleum refining industry, pharmacy industry, and so on. The algorithm need not identify the coupling functions of process systems, is simple and convenient. It will save resource, improve the performance of MIMO control systems, produce some economic benefits.