L-M贝叶斯正则化BP神经网络在红外CO_2传感器的应用

针对温度会影响红外CO_2传感器的输出电压,造成对CO_2的浓度检测误差较大的问题,提出了一种基于L-M贝叶斯正则化BP神经网络的温度补偿方法。实验中将传感器输出电压比和温度作为神经网络的输入,CO_2浓度作为神经网络的输出,并通过L-M算法和贝叶斯正则化对神经网络进行优化。经过实验仿真证明,在温度补偿后红外CO_2传感器测量输出的浓度值最大相对误差为4.557 8%,具有较高的精确度。因此L-M贝叶斯正则化BP神经网络能对红外CO_2传感器进行有效的温度补偿,可为相关红外传感器仪器的改进提供参考。

Application of BP neural network with L-M Bayesian regularization in infrared CO2 sensor

Aiming at the influence of temperature on the output voltage of infrared CO₂ sensor and the detection error of CO₂ concentration, a temperature compensation method based on L-M Bayes regularization BP neural network is proposed. The output voltage ratio of the infrared CO₂ sensor and temperature are taken as input of neural network, CO₂ concentration is used as output of neural network, and neural network is optimized by L-M algorithm. The experimental simulation shows that the maximum relative error of the measured output is 4.557 8% after temperature compensation, which has high accuracy. Therefore, the L-M Bayesian regularization BP neural network can effectively compensate the temperature of the infrared CO₂ sensor, which provides a reference for the improvement of related infrared sensor instruments.