基于激光雷达探测技术的PM_(2.5)浓度辨识研究

PM_(2.5) Concentration Identification Based on Lidar Detection

针对颗粒物浓度的大气分布难以测量的问题,采用532 nm激光雷达,对淮南地区2016年6月1日至12月31日进行连续观测。利用大气边界层高度、气溶胶光学厚度、温度、相对湿度、风速、能见度和实测的颗粒物浓度建立回归预测模型,实现了对颗粒物浓度的辨识研究。由于传统的反向传播(BP)神经网络易陷入局部极小,依据数据特点采用基于遗传算法的反向传播(GA-BP)神经网络进行研究,利用遗传算法寻找最优的权值和阈值,以平衡全局与局部的矛盾。通过两个回归模型的比较,可知GA-BP方法明显优于BP方法,BP方法的测试集的相关指数R^(2)是0.623,平均预测误差是24.692μg/m^(3);GA-BP方法的测试集的相关指数R^(2)是0.899,平均预测误差是7.122μg/m^(3)。由此说明激光雷达可以有效地监测大气颗粒物的分布,并为淮南地区的颗粒物监测提供数据支持和参考依据。

For the difficulty in measuring the distribution characteristics of PM_(2.5) concentration in the atmosphere,we used 532 nm lidar to continuously observe the Huainan area from June 1st to December 31st,2016.A regression prediction model was established concerning the atmospheric boundary layer height,aerosol optical depth,temperature,relative humidity,wind speed,visibility,and measured PM_(2.5) concentration to identify the PM_(2.5) concentration.Since the traditional backpropagation neural network(BP)was prone to the local minimum,we adopted a genetic algorithm-based backpropagation neural network(GA-BP)according to the data characteristics and applied the genetic algorithm to finding the optimal weights and thresholds,balancing global and local contradictions.A comparison of the two regression models demonstrates that the GA-BP method is significantly better than the BP method.The correlation index R2of the test set and the mean forecast error are respectively 0.623 and 24.692μg/m^(3) for the BP method,and 0.899 and 7.122μg/m^(3) for the GA-BP method.These results indicate that lidar can effectively monitor the PM_(2.5) distribution in the atmosphere and provide data support and reference for the monitoring of atmospheric PM_(2.5) in the Huainan area.