BP神经网络在室内甲醛定量分析中的应用

针对室内低浓度甲醛气体的定量分析中,反向传播(BP)算法初始权重敏感性、容易陷入局部极值等问题,以遗传算法优化BP网络,对浓度范围在(0.002~0.06)×10 -6的30组不同浓度的甲醛气体进行定量分析.通过对室内气体中的甲醛气体的初始数据进行优化,将优化的权值阈值代入BP网络,进行浓度的回归分析,并与BP神经网络模型回归效果对比,结果表明:遗传算法优化BP网络方法运行时间约为BP网络的1/2,且预测精度明显高于BP网络.相较于BP网络,遗传算法与BP网络结合更适合处理甲醛气体定量分析问题.

Application of BP neural network in quantitative analysis of indoor formaldehyde

Aiming at problems of quantitative analysis of low concentration formaldehyde gas in indoor,it is difficult to employ the initial weight sensitivity,easily fall into local extremum of BP algorithm. BP network optimized by genetic algorithm(GA)is improved.30 groups of formaldehyde ranging of(0.002~0.06)×10 -6is quantitatively analyzed.Optimized weight value threshold is substituted into the BP network,and the regression analysis is carried out.Experimental results show that the running time of the optimized BP algorithm is about 1/2 of the BP network,and its prediction precision is higher than BP network. Compared with BP network,the combination of GA and BP network is more suitable for dealing with quantitative analysis of formaldehyde gas.