光学学报, 2017, 37 (5): 0501001, 网络出版: 2017-05-05
两种估算近海面大气光学湍流强度方法的比较
Comparison of Two Approaches for Estimating Atmospheric Optical Turbulence Intensity near Sea
大气光学 折射率结构常数 统计学习理论 湍流估算模型 atmospheric optics refractive structural constant statistical learning theory turbulence estimation model
摘要
选取随船观测的三亚地区2016-01-06至2016-01-09连续4天的折射率结构常数C2n及温度、风速、相对湿度三种常规气象参数, 基于后向传播神经网络和逐步回归理论, 分别建立两种模型并对C2n进行了连续3天的估算。结果显示, 两种模型估算的结果在变化趋势及量级上均符合近海面光学湍流的一般特征和变化规律, 并且可以表现出C2n的基本日变化特征, 整体相关系数分别为0.8661和0.8496。选取了平均绝对误差、平均相对误差、均方根方差以及相关系数等统计量来衡量估算结果。分析表明, 两种模型均能准确地估算出近海面的C2n, 但在夜间弱湍流发生时估算值略高于测量值。为进一步提高估算的准确度, 需要改进模式在夜间的估算效果。
Abstract
A refractive index structural constant C2n and three kinds of conventional meteorology parameters (temperature, wind speed and relative humidity) are chosen from ship-based measurement in Sanya from 2016-01-06 to 2016-01-09. Two models are established based on the backward propagation neural network and the stepwise regression theory, and three-day estimation of C2n is carried out. The results show that the variation tendency and magnitude of the results estimated by two models are in accord with general characteristics and change rule of the optical turbulence near the sea, and these results demonstrate the fundamental diurnal variation of C2n. Overall correlation coefficients are 0.8661 and 0.8496. Statistical variables of mean absolute error, mean relative error, root mean square variance and relative coefficient are used to evaluate the estimation results. Further analysis shows that both the two models can calculate C2n near the sea precisely. However, when weak turbulence occurs at night, the estimation results are slightly higher than measurement results. To further improve the estimation accuracy, the estimation effect during nighttime should be improved.
吕洁, 朱文越, 蔡俊, 青春. 两种估算近海面大气光学湍流强度方法的比较[J]. 光学学报, 2017, 37(5): 0501001. Lü Jie, Zhu Wenyue, Cai Jun, Qing Chun. Comparison of Two Approaches for Estimating Atmospheric Optical Turbulence Intensity near Sea[J]. Acta Optica Sinica, 2017, 37(5): 0501001.