Research On Intelligent Controller Optimized By Genetic Algorithm

PID controller is still widely used in industrial control fields, because of its clear principle, simple structure, and easily implementation. For adapting to different objects, the key to the application of PID controller is the selection of appropriate parameters. But for an object with time-varying, large time delay characteristics, the PID controller's effect is limited. As lacking the capability of justifying the parameters online, it is difficult for the PID controller to adapt to a changing environment. Many researchers have been working hard to improve this problem from when the PID algorithm has been proposed.Artificial Neural Network (ANN) is an imitation of the intelligence by the point of physiological. It has a high capacity of learning and adaptive, can approximate any function to arbitrary accuracy, and complete the simulation of the system. The Genetic algorithm is a simulation of natural biological evolution, which has a strong ability of global optimization. These two algorithms are more intelligent method of current research. The idea of combining these two methods with the conventional PID controller to be a intelligent controller with the abilities of parameter auto-tuning and adaptive for the requirements of the complex environment, has a high practical significance of improving the control effect.The work of this paper mainly consists of three parts:First, study the conventional PID algorithm and the physical meaning of each parameter deeply including their' effect on control. The related improvements on all aspects of PID controller and the common methods of adjusting the parameters will be analysis. Second, focus on the research of the PID intelligent algorithm combined with BP Neural Network. Justify the PID controller's three parameters by the ability of self-learning of BP Neural Network. As based on the gradient descent, the BP algorithm will get a local minimum with a certain probability, then, can not reach the global optimum. Therefore, further discuss the relevant improvement of BP algorithm.Third, because of its unique operation, the Genetic Algorithm could converge to the global optimum with a probability 1. With a strong ability of global optimization, it is very suitable for the optimization of a class of implicit functions, such as Artificial Neural Network. This paper has discussed the combination method of Genetic Algorithm and Neural Network, and uses this method for modifying PID controller's parameters online.