| Neural network, as one of modern information process technologies, has some special advantages in many applications. Neural network PID control technology in which plays a very important role, also has a very high value of research and application. This paper discusses such issues due to current algorithm of PID control of neural network being drawbacks of local extreme minimum.The main work of the thesis is processed on the following facets:1. The simulation analyses nonlinear system identification of RBF neural network,recognizes that neurons provide Jacobian information to the controlled object and subsequently achieves online adjustment of PID control parameters.2. BP algorithm adjusts kinds of parameters as the intial weights of neural network those have been optimized by genetic algorithm.The simulation shows that control accuracy optimized by genetic algorithm is better than that has not done, and proves the feasiblity of combination of evolutional algorithm and RBF neural network.3. Differential evolutional algorithm, new evolutional algorithm has caught comprehensive attention of scholars of all countries. A mass of researches show,differential evolutional algorithm has good local optimization ability, it gets extensively application in the optimization of nonlinear systems. As the simulation gives a in-depth study of principles, structures and algorithms of differential evolutional algorithm and proves the feasiblity of combination of differential evolutional algorithm and RBF neural network, which solves the disadvantage as a result of imprecise assignment of intial weight of neural network.4.Since the BP algorithm readily gives local extremum,it is recommended that RBF neural network trained by ISA algorithm could blend with DE to compose hybird algorithm thereby optimizing the PID control of RBF network and being applicable to the nonlinear time-varying control system.A logical next step is to justify the effectiviness of this hybrid algorithm. |
PDF Full Text Request