基于BP神经网络的滑坡监测多源异构数据融合算法研究

针对滑坡监测中的多源异构数据融合问题，论文提出了一种基于BP神经网络的多源异构监测数据融合算法。该算法将影响滑坡变形的温度、湿度、风力、云量、单日降水量和累计降水量等多环境因子变量作为输入变量，以滑坡位移变化量数据作为期望输出数据，并利用各环境因子变量和滑坡位移变化量的相关性及显著性进行环境因子变量筛选，以提高算法的预测精度。论文采用甘肃省永靖县黑方台党川滑坡的实测数据进行了试验，结果表明:反向传播（Back-Propagation，BP）神经网络数据融合算法适用于具有多源异构监测数据的滑坡变形预测；在进行环境变量因子筛选后，BP神经网络数据融合算法的决定系数达到0.985，均方根误差（RMSE）达到0.4787 mm，从而有效提高了变形预测结果的精度。 

Research on multi-source heterogeneous data fusion algorithm of landslide monitoring based on BP neural network

Aiming at the multi-source heterogeneous data fusion problem of landslide monitoring, a multi-source heterogeneous monitoring data fusion algorithm based on BP neural network is proposed in this paper. The temperature, humidity, wind power, cloudiness, precipitation and accumulated precipitation which affect the landslide deformation are taken as the input variables, and the landslide displacement changes data are taken as the expected output data in this algorithm. And the prediction accuracy of this algorithm can be effectively improved by filtering the environmental factor variables with calculating the correlation and significance of the environmental factor variables and the landslide displacement changes. This algorithm is verified by the monitoring data of the Dangchuan landslide in Heifangtai, Yongjing County, Gansu Province. The results show that the BP neural network data fusion algorithm can be used in the landslide displacement prediction with multi-source heterogeneous monitoring data. After the environmental factor variable filtering, the determination coefficient of the BP neural network data fusion algorithm can achieve 0.985 and the RMSE can achieve 0.4787 mm. Thus the accuracy of deformation prediction can be effectively improved.