The Research of New Black Box Control Method Based on Conjugate Gradient Algorithm

With the development of artificial neutral network and control science, black box control has become one of the most popular topics for the researchers because of its good performance of the self-adaptivity, robustness and antidisturbance in recent years. Since there are lots of drawbacks for the BP neutral networks such as low converging speed and uncertainly of network structure and weight factors. This paper develops a new modified F-R algorithm to improve converging speed of back propagation neutral network and tries to eliminate the bad effect to the whole control system caused by uncertainty. The topology structure and weight factor of the neutral network are optimized by using GA (Genetic Algorithm) offline. This paper introduces servo control system, network optimization algorithm, gradient descent algorithm, and modified Fletcher- Reeves algorithm. The black box algorithm is programed in MATLAB and simulated in the Simulink for control system. During the simulating experiment, the load disturbance is added to test the capability to withstand the disturbance. The results show that the modified Fletcher-Reeves algorithm has the better performance in the response time, overshooting and antidisturbance ability compared with other two methods. In the end, the experiment is carried out based on the successful simulation. The control program is finished in LabVIEW and applied to the servo-control systems of EMA (Electron-mechanic Actuator). The results indicate the response of the system has the better stability and rapidity, which can meet the requirements of engineering application greatly.