BP Neural Network Optimization And Research

Artificial neural network is a hot research field in recent years; it is an important part of research in artificial intelligence. It has become the focus of neuroscience, computer science, cognitive science, mathematics and physics and other disciplines. Its application areas include:classification, prediction, and pattern recognition, signal processing and image processing, and constantly expand. In the beginning, Optimized and improved neural network is always an important research contents in neural network theory research and application field, the feed-forward neural network has not been a very satisfactory solution in particular. In this paper, BP neural network with the broadest applications and most representative in feed-forward neural network is an research object, do some research on BP neural network, propose a kind of improved model-Hybrid BP neural network. This model is proposed that because BP neural network has complex network structure and low classification ability in complex sample classification.BP neural network has been trap into local mina value easily and has slow convergence problem. This paper proposed that hybrid BP neural network model and applied this model in complex sample classification. The model constructs network structure by analyzing the correlation between samples of properties. This model uses the principal component analysis (PCA) to reduce the dimensions of the sample and uses the bee colony algorithm (ABC) to optimize the weights of the network. The former reduces the relevance of the data sample, using principal component to represent the original data, reduce network input layer number of neurons and the computational complexity; the latter solve the neural network weights on the selected the random problems and avoid weight selected as a result of random networks of falling into local minima.To verify the effectiveness of hybrid BP neural network, this paper compared with the traditional BP neural network and GA-BP neural network in solving classification of complex sample. The simulation results show that this method can not only improve the BP neural network's ability in the data classification, and using hybrid BP neural network to solve complex classification is feasible and effect is superior to GA-BP neural network and BP neural network.