The Optimization Research Of Feedforward Neural Network Based On Genetic Algorithms

Along with the recovery of the artificial neural network in the 1980s, a worldwide boom of the neural network research has being coming. Currently, the BP neural network is one of the most widely used neural networks. The topology of BP neural network is a multilayer feed forward networks. The theory of BP neural network and BP learning algorithm are mature, and the BP network based on error back propagation algorithm has strong nonlinear mapping, generalization, fault-tolerant ability. The practice proved that BP network can solve a lot of practical problems. But in practice, BP algorithm also revealed some weaknesses, such as the fixed adopted and the number of the training, makes the study low efficiency, slow convergence speed; the selection of the hidden layer and hidden-layer points lack of the theoretical guidance; standard gradient descent method with continuously reduced based on error and performance index function has been used, making the ability of global network search poor, easy to form local minimal and can't get the global optimal; forgetting old sample when training learning new samples, and etc.The appearance of genetic algorithms makes the neural network training having new look. Using genetic algorithms to optimize the neural network can make the neural network having since evolution and adaptive ability, thus constructed the evolution neural network. Combining the genetic algorithms and the BP neural network has a great significance. Due to the genetic algorithms can converge to the global optimal solution, it can not only play the BP neural network generalization ability, but also make BP neural network has quick convergence and strong learning ability. The combined of the genetic algorithms and the BP neural network has two aspects: one aspect is using the genetic algorithms to optimize the BP neural network's connection weights between each layer value; the other is using genetic algorithms to optimize the topology of BP neural network.The paper systematically studied the artificial neural network, especially BP neural network and genetic algorithm, analyzed the neural network's defect and its causes. Based on this basis, applying genetic algorithm optimized the BP neural network's connection weights and the topological structure. In the optimization process, this paper applied the genetic algorithm with the coding skills suitable for neural network according to the actual code, and specially designed the variation operating plan according to the coding method of genetic .After optimization , the BP neural network has been applied in colon cancer gene expression profile data analysis, and the results indicate that the optimized BP neural network has faster convergence and more strong learning ability than the traditional BP neural network. The optimized network overcame the BP neural network's some defects, makes the BP neural network application more perfect in the practice.