GA-BP神经网络与BP神经网络性能比较

Performance Comparison between GA-BP Neural Network and BP Neural Network

目的在MATLAB软件上拟合BP与GA-BP神经网络数据,比较传统BP与GA-BP神经网络拟合数据的效果,并确定网络模型。方法利用计生、卫生部门联合开展VCT服务影响因素分析结果的资料,研究采用遗传算法优化BP神经网络的初始权值和阈值,并通过"试错法"确定隐含层神经元数,比较传统BP神经网络与GA-BP神经网络拟合数据的效果。结果 GA-BP神经网络相对于BP神经网络拟合数据迭代步数更少、能更快地达到预设目标;在R2和调整R2无统计学差异的前提下,当隐含层神经元为15时,BP神经网络和GA-BP神经网络均比较稳定,GA-BP拟合效果更好。结论 GA-BP神经网络建模稳定性高,GA-BP神经网络较BP神经网络能达到预设目标的次数更多,能达到全局最优,表明遗传算法优化BP神经网络具有可行性。

Objective To carry on fitting data by using BP and GA-BP neural network on the MATLAB software. To compare their fitting effects and determine the network model. Methods Use the mate- rial of influence factor analysis results on Department of Health and depart- ment in charge of family planning carrying Health AIDS Voluntary Counse- ling and Testing（VCT） services jointly. This paper used genetic algorithm to optimize initial weights and thresholds, and determined hidden neurons through the ＂trial and error＂. It can compare their fitting effects of the conditional BP neural network and GA-BP neural network. Results The iterative step of fitting data of GA-BP neural net- work was less than that of the BP neural network, and it can reach the in- tended target faster ; under the premise of R2 and the adjust R2 no statistical difference, BP neural network and GA-BP neural network were relatively stable when hidden layer neurons is 15. GA-BP fitting effect was better. Conclusion GA-BP neural network modeling stability is great. GA-BP neural network can reach the rate of the intended target more than BP neural network, and it can reach the global optimum. It indicates that the genetic algorithm to optimize the BP neural network is feasible.