基于机器学习的高程异常建模算法及其水电工程的应用  

Elevation Anomaly Modeling Based on Machine Learning Algorithm and Its Application in Hydropower Engineering

作　　者：潘国俊[1] PAN Guojun(Geological Exploration Technology Institute of Jiangsu Province,Nanjing 210049,Jiangsu,China)

机构地区：[1]江苏省地质勘查技术院,江苏南京210049

出　　处：《水力发电》2023年第6期28-33,共6页Water Power

摘　　要：某水电工程测区受地形复杂、植被覆盖率高等因素的影响,水准测量实施困难,采用循环神经网络RNN、反向传播BP神经网络和径向基函数RBF神经网络等3种机器学习算法,分别对试验区域高程异常进行拟合与建模,并将结果与二次曲面拟合方法进行对比。内符合与外符合精度对比发现,机器学习算法建立的高程异常模型精度高、残差小。3种方法中,径向基函数RBF神经网络更适用于研究区域的高程异常建模。此外,基于相同的机器学习算法,研究随机选取部分拟合点高程异常建模的精度发现,公共点分布更加均匀时,拟合效果更好。研究成果为复杂地形条件下水电工程项目高程异常建模提供参考。Due to complex terrain and high vegetation coverage rate,it is difficult to measure the leveling in a hydropower project survey area.Therefore,three machine learning algorithms of recurrent neural network(RNN),back-propagation(BP)neural network and radial basis function(RBF)neural network are used to fit and model the elevation anomalies in the test area,and the results are compared with the quadric surface fitting method.By comparing the accuracy of internal and external coincidence,it is found that the elevation anomaly model established by machine learning algorithm has high accuracy and small residual error.Among the three methods,the radial basis function(RBF)neural network is more suitable for the elevation anomaly modeling in the study area.In addition,based on RBF neural network algorithm,the accuracy of elevation anomaly modeling is also studied when some fitting points are randomly selected,and the results shown that the fitting effect is better when the distribution of common points is more uniform.The research provides a reference for elevation anomaly modeling of hydropower projects under complex terrain conditions.

关 键 词：水电工程 GNSS高程异常 机器学习算法 拟合 建模 循环神经网络 反向传播神经网络 径向基函数神经网络 

分 类 号：TV221.1[水利工程—水工结构工程]