Abstract：Because of the varying concentration of atmospheric PM2.5 have strong nonlinear characteristics, traditional soft sensor methods are difficult to make accurate measuring and monitoring. According to traditional BP neural network is easy to fall into local minimum, BP neural network is combined with genetic algorithm to establish the GA-BP neural network soft sensor model. The model is applied to the monitoring of the atmospheric concentration of PM2.5, and compared with the results of the monitoring of the traditional BP neural network model, the results show that the genetic algorithm optimization model has a better non-linear fitting ability and higher monitoring accuracy.
［1］郭清彬,程学丰,侯辉,等.冬季大气中PM10和PM2.5污染特征及形貌分析［J］.中国环境监测,2010,26(4):55-58.
［2］Baker K R, Foley K M. A nonlinear regression model estimating single source concentrations of primary and secondarily formed PM2.5［J］.Atmospheric Environment,2011,45(22):3758-3767.
［3］李松,罗勇,张铭锐.遗传算法优化BP神经网络的混沌时间序列预测［J］.计算机工程与应用,2011,47(29):52-55.
［4］傅荟璇,赵红.MATLAB神经网络应用设计［M］.北京：机械工业出版社,2010.
［5］武常芳.基于B-P神经网络方法的西安市四季PM10浓度预报模型研究［D］.西安：西安建筑科技大学,2007.
［6］Saad E W,Wunsch D C.Neural network explanation using inversion［J］.Neural Networks,2007,20(1):78-93.
［7］MATLAB中文论坛.MATLAB神经网络30个案例分析［M］.北京：北京航空航天大学出版社,2010.
［8］刘艳华.基于遗传神经网络的西安市采暖季PM10污染浓度预报研究［D］.西安：西安建筑科技大学,2009.
［9］Ashena R, Moghadasi J. Bottom hole pressure estimation using evolved neural networks by real coded ant colony optimization and genetic algorithm［J］. Journal of Petroleum Science and Engineering,2011,77(3-4):375-385.
［10］赵宏,刘爱霞,王恺,等.基于GA-ANN改进的空气质量预测模型［J］.环境科学研究,2009, 22(11):1276-1281
2014, Vol. 35 
