基于自适应遗传BP算法的混合气体定量检测研究

摘要

An optimization model with the adaptive genetic algorithm and the improved BP neural network was presented for the quantitative detection of gas mixtures.To overcome the disadvantages of BP-ANN such as slowly searching rate and partially leading to minimum,the adaptive genetic algorithm was used to get better initial weights and thresholds of the BP network in the early stage,The network was trained by the error back propagation method.A three-layer 7 × 18 × 3 BP neural network was designed for a group of gas mixtures with five samples.The results show that the convergence speed and the learn precision of adaptive genetic algorithm optimizing BP neural network are better than that of the adding momentum BP algorithm.And the application to the recognition of gas mixtures is reliable and the method can improve the detection efficiency of gas mixtures,and the time consumption is reduced to 1/3.%针对混合气体检测问题,利用误差反向传播(BP)算法和遗传算法,提出了用自适应遗传算法优化BP神经网络的方法来实现定量检测.即利用遗传算法的全局搜索能力,对神经网络连接权值和阈值进行优化,再以优化后的初值作为BP神经网络的初始连接权值和阈值,最后用附加“动量项”的误差反向传播算法训练BP网络.设计了一个结构为7×18 ×3的3层BP网络用于一组含有5个样本的混合气体识别试验.结果表明:将自适应遗传神经网络算法应用于混合气体定量识别的训练中,收敛速度比附加“动量项”BP算法要快,而且学习精度更高,识别效率也提高了2/3.