基于PS-InSAR技术和遗传神经网络算法的矿区地表沉降监测与预计

针对传统雷达干涉测量技术(D-InSAR)易受大气相位延迟和失相关的影响以及传统BP算法依赖于初始权值和阈值问题.本文采用了(PS-InSAR)技术对矿区地表沉降进行了监测,并提出采用遗传算法(GA)对神经网络(BP)算法的初始权值和阈值进行筛选.首先利用PS-InSAR技术获取矿区地表沉降范围和沉降值,然后将其部分结果作为遗传神经网络(GA-BP)算法的训练样本建立预测模型参数.选取宿州市矿区19景Sentinel-1A雷达数据进行实验分析,结果表明,PS-InSAR技术能够很好监测矿区地表沉降,最大沉降速率为45 mm/a.分别取训练样本数为1000、2000、3000和4000利用GA-BP算法对矿区地表沉降进行预测,得到最大残差分别为6.8 mm、0.44 mm、0.36 mm、0.28 mm;均方误差分别为3.85 mm、3.26 mm、2.98 mm、1.61 mm,表明本文提出的GA-BP算法能有效预测矿区地表沉降,并且在训练样本数量较多时预测效果和预测性能较好.

Surface subsidence monitoring and prediction based on PS-InSAR technology and genetic neural network algorithm

The traditional radar interferometry(D-InsSAR)is susceptible to the influence of atmospheric phase delay and loss correlation,and the traditional BP algorithm relies on the initial weight and threshold.In this paper,the PS-InSAR technique was used to monitor the surface subsidence in the mining area,and the Genetic Algorithm(GA)was proposed to screen the initial weight and threshold of the neural network(BP)algorithm.Firstly,the PS-InSAR technique was used to obtain the surface subsidence range and subsidence value of the mining area,and then some of the results were taken as the training samples of the genetic neural network(GA-BP)algorithm to establish the prediction model parameters.Sentinel-1a radar data of 19 scenes in Suzhou mining area were selected for experimental analysis.The results showed that PS-InSAR technology could well monitor surface subsidence of mining area with a maximum subsidence rate of 45 mm/a.The number of training samples was respectively 1000,2000,3000 and 4000.GA-BP was used to predict the surface subsidence of the mining area,and the maximum residual was 6.8 mm,0.44 mm,0.36 mm and 0.28 mm,respectively.The mean square error was 3.85 mm,3.26 mm,2.98 mm,and 1.61 mm,respectively,indicating that GA-BP algorithm could effectively predict surface subsidence in mining areas,and had better prediction effect and prediction performance when the number of training samples was large.