【摘要】 封闭鸡舍环境中存在多种有毒气体,常用的半导体气体传感器选择性差、受气体交叉干扰的影响大,导致测量误差大,不能满足检测需求。针对这个问题,采用经典数据融合方式中的最小二乘法和现代数据融合方式中的BP神经网络的4种训练函数,分别对多气体传感器输出信号进行仿真训练分析,并利用均方误差和迭代次数来评价仿真的性能。仿真和实验结果表明,有弹回的BP算法训练的网络性能最优,可有效地降低测量误差,平均相对误差和最大相对误差均在1%以内,满足封闭鸡舍环境检测需求。更多还原

【Abstract】 There are a variety of toxic gases in the closed henhouse environment,but the poor selectivity of commonly used semiconductor gas sensor and the great effect of the cross coupling cause too large measurement error to meet the demand of measurement. Aiming at these deficiencies,the least squares method in the classical data fusion method and the 4 training functions of the BP neural network in the modern data fusion method to simulate and analyze the output signal of themulti-gas sensors are used to improve the measurement accuracy. The training square error and network iteration are adopted to evaluate the performance of the model. Simulation results show that the network performance trained by the rebound BP algorithm is optimal to reduce the measurement error effectively. The average relative error and the maximum relative error are limited within 1%to meet the requirements of environmental testing.更多还原