Optimization Research And Application Of BP Neural Network

With the development of science and technology, people have been trying to leta machine with the human thinking and process the information from the outsideworld intelligently, and artificial neural network was born in this background.Artificial neural network is a mathematical model of the human brain simulationmode. By simulating the human brain system for information storage, processing andhandling, its application field is very extensive. BP(Back Propagation) neuralnetwork is the most widely used artificial neural network, but due to some of its own,such as easy to fall into local minimum, the existence of the network structure isdifficult to determine the inherent characteristics, the scope of its application isstill limited and restricted to a certain extent.This article from the perspective of the genetic algorithm and BP neural networkconvergence to optimize the network, the main points are as follows:First of all, the relevant knowledge of biological neurons and artificial neuronsare introduced, and then the most widely used BP neural network structure, principleand implementation are explained, and use mathematical formulas to algorithms for adetailed derivation, and finally through a function intended to together examples toanalyze the performance of BP network. Simulation results show that the presence ofthe standard BP algorithm, such as the limitations of slow convergence and networktraining function, the number of intermediate nodes in the hidden layer, the initialparameters can affect network performance.Secondly, we study the genetic algorithm from the basic operations,characteristics and principles and other aspects. Genetic algorithm is a bionicalgorithm based on evolutionary theory, and it is based on the fitness functionconverted from the objective function to conduct selection, crossover and mutation onpopulation.Thirdly, from the perspective of the genetic algorithm and BP neural networkconvergence, we use the genetic algorithms to optimize the BP network, and theoptimization process includes the parameters and the network topology. When the structure of BP network optimization using genetic algorithms, we use the encodingmethod is simple, easy genetic manipulation of binary coding method to achieve.When the weights and threshold of the network to be optimized, we used thereal-coded method, which code length is shorter and performance is more intuitive.The final function fitting simulation experiments show that the performance ofoptimized BP network has been improved.At the end of this paper, we predict the intersection traffic flow data of a road inGuangzhou and the results demonstrate the superiority of optimized BP network.