| With the development of information technology,people have more and more application requirements for artificial intelligence in their lives.Artificial neural networks,especially BP neural networks,have received great attention as tools for realizing artificial intelligence to satisfy these needs.It can establish a simplified model to simulate the decision making process of biological neural networks.The classification problem,as the basic means in the process of human learning,is the most basic way of dealing with information in people's daily life,and it provides the basis for judgment for various types of decision-making.The use of artificial neural networks to solve classification problems has been the topic research of many researchers.As a classical feedforward neural network,BP neural network is the most widely used neural network for classification problems.However,BP neural network still has a large optimization space as a traditional neural network algorithm.Its defects are mainly manifested in the fact that it is easy to fall into the local minimum value during the weight training process,the error convergence speed is slow,and the performance is weak when classifying the fuzzy information.Based on BP neural network algorithm,this paper studies the optimization of classification problem for BP neural network model from performance optimization and fuzzy information processing.Mainly completed the following work:(1)By studying and researching the difference between the algorithm of simulated annealing algorithm and gradient descent algorithm for BP neural network weight iteration,the paper proposed the plan that adjusts the weight of BP neural network by Metropolis criterion in simulated annealing algorithm,so that BP neural network classifier model can reach the minimum error faster and more accurately.(2)A classification method combining fuzzy theory and artificial neural network is realized.This classification method improves the ability of BP neural network classifier to process fuzzy information to some extent.(3)The system design of BP neural network classifier algorithm based on fuzzy theory and simulated annealing algorithm is implemented on the visual development platform.The classification experiment is carried out in this system and the experimental results are analyzed.After comparing the actual data sets of multiple UCIs in the system,it is found that the improved neural network classification has certain advantages compared with the traditional BP neural network algorithm.It not only ensures the fuzzy data is using the neural network to simulate nonlinear problems.The classification accuracy and classification stability of the time can also make the network model converge more efficiently to the error extreme point during training and learning. |
