基于GPS和神经网络的滑坡位移监测算法

针对使用GPS接收机进行滑坡位移监测时，单点GPS定位误差较大的问题，提出了一种基于GPS和神经网络的滑坡位移监测算法。GPS接收机在滑坡发生之前和滑坡发生之后测得的定位数据耦合在一起，不是线性可分的。采用具有非线性可分特性的神经网络，把耦合在一起的定位数据分成两类，一类属于未滑坡的GPS数据，另一类属于发生滑坡的GPS数据，避免了对GPS定位误差这一非线性非高斯问题进行准确建模的过程。利用GPS接收机测得的样本训练集训练神经网络，用训练后的神经网络模型来验证测试集的分类效果。实测实验结果表明，对于低精度的GPS接收机，当滑坡程度分别达到3米、5米、8米时，训练样本分类正确率分别是95.85%、99.23%、99.99%，测试样本分类正确率分别是82.94%、89.44%、91.05%，说明所提出的算法适用于较大程度的滑坡。

Landslide Displacement Monitoring Algorithm Based on GPS and Neural Network

Aiming at the problem of single-point GPS positioning error when using GPS receivers for landslide displacement monitoring, a landslide displacement monitoring algorithm based on GPS and neural network is proposed. Before the landslide and landslide occurred, the positioning data measured by GPS receivers are coupled together and are not linearly separable. The neural network with non-linear separable features is used to divide the coupled positioning data into two classifications, one belongs to the non-landslide GPS data and the other belongs to the landslide GPS data, which avoids the accurate modeling of GPS nonlinear and non-Gaussian positioning error. The sample training set measured by the GPS receiver is used to train the neural network, and the trained neural network model is used to verify the classification effect of the test set. The experimental results show that for the low-precision GPS receivers, when the landslide reaches 3 meters, 5 meters and 8 meters respectively, the correct rates of training samples classification are 95.85%, 99.23% and 99.99% respectively, and the correct rates of testing samples classification are 82.94%, 89.44% and 91.05% respectively, indicating that the proposed algorithm is suitable for a greater degree of landslide.