混沌遗传神经网络在空气质量预测中的应用  

Application of chaotic genetic neural network to the air quality forecast

作　　者：孟栋[1] 樊重俊[1] 李旭东[1] 卜宾宾 

机构地区：[1]上海理工大学管理学院,上海200093

出　　处：《安全与环境学报》2014年第4期246-250,共5页Journal of Safety and Environment

基　　金：上海市教育委员会科研创新重点项目(14ZZ131)

摘　　要：为了提高对空气质量预测的准确性,提出了一种基于混沌遗传算法(CGA)的BP神经网络改进方法。BP神经网络是目前应用最广泛的神经网络,但存在收敛速度慢和易陷入极小值的缺陷。该改进算法的基本思想是用混沌遗传算法优化BP神经网络的初始权值和阈值。混沌遗传算法结合了混沌运动的遍历性和遗传算法的反演性。将混沌变量加入遗传算法中,进一步提高了遗传算法的全局搜索能力和收敛速度;将混沌遗传算法优化后得到的最优解作为BP神经网络的初始权值和阈值。利用改进后的CGA-BP算法进行空气质量预测,结果表明,该方法对空气质量的预测效果明显好于单纯使用BP神经网络的预测效果。This paper is aimed to propose an improved method for the BP neural network control based on the chaos genetic algorithm （CGA） in hoping to increase the accuracy of the air quality forecast for the method is one of the most widely used ones in the BP neural network control. The proposed optimized process is supposed to con tain the chaos generation of the initial population, the fitness func tion, the selection operation, the crossover operation, the mutation operation along with the individual chaos control optimization. How ever, among the numerous CGA methods we would like to choose the BP neural network control, though it is disadvantageous for its ergod icity of the chaotic motion and the inversion of the genetic algorithm （GA）. The fundamental starting idea for the improvement we suggest is to better the aforementioned weak points and to optimize the initialweights and the threshold of BP network. The socalled CGA method can be thought to join both the ergodicity of chaotic motion and the inversion of genetic algorithm （GA） into its own in the process of generating the initial population plus the elements of chaos, which help to expand the scope of traversal to the entire scope of variables to get rid of the results involved in the local optimum and premature rush during the process and heighten the global searching ability and the convergence speed further. In the end, the optimal solution can be regarded as the initial weights and threshold of BP network. As the result, the latest two days＇ air quality, the air quality index （AQI）, and the six basic monitoring indexes, the meteorological conditions forecasted for yesterday and the day can all be the network forecast predicted results, which can then be used into the training jobs in the air quality forecast and evaluation. What is more, fitness should he decided as the result of calculating the network feedback, then dealt it in order of selection, copy, crossover and mutation operation. Ac cording to the fitness function, it would be po

关 键 词：环境工程学 空气质量 预测 混沌 遗传算法 BP神经网络 

分 类 号：X831[环境科学与工程—环境工程]