基于BP神经网络的红外透过率计算

Calculation of infrared transmittances based on BP neural network

针对实时条件下中红外波段平均大气透过率的计算,提出了一种基于贝叶斯正则化BP神经网络的方法。利用BP神经网络良好的非线性拟合特点,建立大气参数与中红外平均透过率之间的关系模型,从而可以准确迅速地得到计算结果。此网络模型是以实测温度、压强、湿度和气溶胶的后向散射系数作为输入向量,分别以水蒸气和CO2吸收透过率、气溶胶散射透过率和大气透过率作为输出。仿真结果表明:在相同的大气参数下,本方法的计算结果与期望值之间的相对误差较小,且远小于经验公式法,验证了本方法的可行性与有效性。因此,本方法对大气透过率的准确地快捷计算提供了有益的借鉴。

In view of calculating atmospheric transmittance in actuality,a method based on back propaga- tion （BP） neural network with Beyesian regularization is proposed. The model of relationship between atmospheric data and infrared transmittances is built by using the characteristic of nonlinear fitting of BP neural network. Therefore,it is fast and precise to calculate infrared transmittances. The input factors in- elude measured temperature,pressure,humidity and back scattering coefficients. The output variables are absorption transmittances of H_2O and CO_2, scattering transmittance and atmospheric transmittance of mid-infrared. Simulations indicate that the relative errors between calculating result and desired value are smaller than those calculated by empirical formula. The conclusion reveals the feasibility and availability of this method that provides a helpful lesson to calculate atmospheric transmittance precisely and con- veniently.