CAJ | 학술논문

针对目前常见的多元有害气体检测问题，搭建了一种基于传感器阵列和GP神经网络相结合的多元有害气体检测系统。该检测系统中采用了GP神经网络算法对传感器阵列的4种混合有害气体的响应信号进行回归分析。为了提高GP神经网络的预测准确性，又利用了粒子群优化（ PSO）算法对GP神经网络的权值与阈值进行了优化。结果显示：通过PSO 优化的GP（ PSO-GP）神经网络预测的平均相对误差小于2%，能够有效解决气体传感器交叉敏感问题。
침대목전상견적다원유해기체검측문제，탑건료일충기우전감기진렬화GP신경망락상결합적다원유해기체검측계통。해검측계통중채용료GP신경망락산법대전감기진렬적4충혼합유해기체적향응신호진행회귀분석。위료제고GP신경망락적예측준학성，우이용료입자군우화（ PSO）산법대GP신경망락적권치여역치진행료우화。결과현시：통과PSO 우화적GP（ PSO-GP）신경망락예측적평균상대오차소우2%，능구유효해결기체전감기교차민감문제。
Aiming at problem of muliple harmful gas detection,a gas detection system is developed by combining sensor array with GP neural network algorithm. This test system uses GP neural network algorithm for regression analysis on response signal of four kinds of mixture harmful gases measured by sensor array. In order to improve prediction accuracy of GP neural network,adopt PSO algorithm to optimize GP neural network weights value and threshold value. The results show that the average relative error predicted by PSO optimized GP neural network ( PSO-GP)is less than 2%,it can effectively solve problem of cross-sensing of gas sensor.