摘要
采用BP神经网络,建立了红外甲烷传感器温度校正模型。以隐层节点数为变量,研究了神经网络规模对训练速度和推广能力的影响。将贝叶斯正规化方法与Levenberg-Marquardt优化算法相结合,提高了神经网络训练的效率和推广能力。所建立的模型网络规模小,运算速度快,可以对温度变化引起的红外甲烷传感器非线性误差进行自动校正。
BP neural network has been applied to the development of the temperature rectification model of IR methane detector. The number of neurons in hidden layer is adopted as an index to the size of the network, whose influence on the learning speed and generalization capacity of the neural network has been thoroughly investigated. Bayesian regularization in combination with Levenberg- Marquardt algorithm has been applied to achieving faster learning speed and well-generalized neural network. The obtained neural network is small in size and exhibits fast computing speed. Therefore, the non-linear error of the IR methane detector in respect of temperature deviation can be automatically rectified.
出处
《煤矿机械》
北大核心
2009年第10期43-45,共3页
Coal Mine Machinery
基金
唐山市科技攻关项目(08160202A-2-2)
关键词
红外甲烷传感器
BP神经网络
温度校正
贝叶斯正规化
infrared methane detector
BP neural network
temperature rectification
Bayesian regularization