Neural Network Based On Improved Genetic Algorithm

The objects controlled become more and more complex with the development of epoch and the progress of technology. The objects controlled have a lot of unclear things, it leads the tradition control methods based on math model is not successful, so the neural network controller is an effective way to solve these problems in intelligent controller. It can be used for uncertain complex system, neural network has the ability of learning, but its inner mechanism isn't clear. It is difficult to present the knowledge. Specifically for the deficiency of neural network and also thinking the optimail algorithms, genetic algorithm has robustness, randomness, global character and merit of being suitable for the parllel processing. The genetic algorithm used to optimize neural network make up deficiency of neural networks.Genetic algorithm is one of the optimal search algorithms that are most widely in use. Today, genetic algorithm has been used in many fields, such as function optimum,model optimum,structure optimum, and so on. Genetic algorithm apts fall into the local optimum, sometimes its convergence speed can not meet for the demands also. How to overcome these limitations has been a hot study project of many scholars and engineers.In view of the SGA is not easy to restrain in the application process, the results often fall into the local optimum, existing the decoding error in encoding methods, the convergence speed is slow etc. The paper proposes a kind of improved genetic algorithm based on the queue selection, and uses it to seek parameter of the intelligent controller superiorly. The selection,crossover and mutation of genetic algorithm which will have the important influence to the ability of seeking in the process of the optimum, and needs corresponding adjustment of the other links to achieve the ideal aim of the seeking superiority of the genetic algorithm. Based on the above reason, this paper put forward the improved genetic algorithm based on the queue selection. The method of queue selection is to queue the individual of the colony after each aim value calculated. Fittness only depends on the position of the queued individual. And the selection possibility of each individual equals to the computation possibility by certain rule of its position.This paper put forward neural network, it influxes the advantage of neural network, gathers learning,association and self-adaptation together. Tradition learning method has disadvantages such as long leaning and slow convergence speed. This paper uses improved genetic algorithm adjusting the weights of neural network and using these optimized parameters as the initial weight values of neural network. Aiming at the structure characters of neural controller, using genetic algorithm training the weights of neural network, the compound controller has good characteristic. So the methods remains the global stochastically searching ability of genetic algorithm, the robustness and self-learning ability of neural network and the extensive mapping ability of neural network and rapid global convergence of genetic algorithm.Using neural control based on improved genetic algorithm in secondary pendulum system, the results of simulation manifests the improved genetic algorithm of queue selection used in looking for the best value of neural network can bring the character of global optimum into play sufficiently and make up the deficiency of long training and slow speed of response in BP.