一种基于随机森林的激光大气传输效应评估方法北大核心CSCD

针对现有的激光大气传输效应评估方法(定标律评估和波动光学仿真评估)较难有良好的模型正则性和应用普适性的困难,提出一种基于随机森林的激光大气传输效应评估方法。该方法首先以烟台某地大气环境数据(温度、风速、湍流强度(C^(2)_(n))等)的随机采样和激光发射参数(传输距离、激光功率等)作为输入数据和多层相位屏模型仿真生成的环围功率(PIB)作为输出数据,然后利用随机森林进行训练和预测。结果表明,随机森林较支持向量机能更好的表征输入与输出间的多元回归关系,预测均方根误差优于0.021%;传输距离、C^(2)_(n)与PIB相关性最强,对PIB影响最大。本方法可为机器学习在激光大气传输效应评估的应用提供更加完善的理论依据,具备一定的应用价值。

A simulation study on atmospheric transmission effectevaluation of laser based on random forest

In view of the difficulties in model regularization and application universality of existing methods(scaling law evaluation and wave optical simulation evaluation)for evaluating laser atmospheric transport effect,a method based on random forest for evaluating laser atmospheric transport effect is proposed.This method firstly takes random sampling of atmospheric environmental data(temperature,wind speed and turbulence intensity(C²_n),etc.)in Yantai and laser parameters(transmission distance,laser power,etc.)as the input data,and multiple phase screen model simulation of turbine power(PIB)as the output data,then with the use of random forests for training and prediction.The results show that the random forest can better represent the multiple regression relationship between input and output than the support vector,and the prediction root mean square error is less than 0.021%.Transmission distance and turbulence intensity C²_nhave the strongest correlation with PIB,and have the greatest influence on PIB.This method can provide a more perfect theoretical basis for the application of machine learning in the evaluation of laser atmospheric transport effect,and has a certain application value.