摘要
针对目前常见的多元有害气体检测问题,搭建了一套基于传感器阵列和集成神经网络相结合的多元有害气体检测系统。为了提高该系统的稳定性和预测精度,提出使用粒子群算法(PSO)优化集成神经网络的权重系数的方法,即利用PSO的全局搜索能力,对该系统的集成神经网络权重系数进行全局优化,再以优化后的权重系数实现多个神经网络的结论结合。该系统对传感器阵列的4种混合有害气体的响应信号进行回归分析。结果显示,该系统PSO算法的集成神经网络预测的平均相对误差小于1%,网络具有更强的稳定性和泛化能力。
Aiming at the common multiple harmful gas detection problem, a gas detection system was developed by combining a sensor array with the integrated neutral-network. In order to improve stability and prediction accuracy of the system, put forward using particle swarm optimization （PSO）to optimize the weight coefficient of integrated neutral-network, namely, utilizing the global search ability of PSO to global optimization of the network weight coefficient, then use the weight coefficient implement multiple conclusions combination of neural network. This system is performing regression analysis on the response signal of four harmful gases mixture measured by sensor array. The results show that the prediction average relative error of the integrated neutral-network optimized by PSO was found to be within 1%, and the network has better stability and generalization ability.
出处
《传感技术学报》
CAS
CSCD
北大核心
2015年第6期938-942,共5页
Chinese Journal of Sensors and Actuators
基金
国家自然科学基金项目(61471210)
浙江省科技厅重大科技专项重点工业项目(2011C16037)
浙江省宁波市科技局自然科学基金项目(2013A610002)
