基于滑坡监测数据的Elman神经网络动态预测  

Elman neural network dynamic prediction based on landslide monitoring data

作　　者：李寻昌[1] 叶君文 李葛 李俊[1] LI Xunchang;YE Junwen;LI Ge;LI Jun(School of Geology Engineering and Geomatics,Chang＇ an University Xi＇ an 110054,China;Shaanxi Nuclear Industry Engineering Investigation Institute Limited Company,Xi＇ an 710054,China)

机构地区：[1]长安大学地质工程与测绘学院,陕西西安710054 [2]陕西核工业工程勘察院有限公司,陕西西安710054

出　　处：《煤田地质与勘探》2018年第3期113-120,126,共9页Coal Geology & Exploration

基　　金：中央高校基本科研业务费资助项目(310826172203;310826161018);陕西省科技统筹创新工程计划项目(2016KTZDSF04-05-04)~~

摘　　要：滑坡在我国是一种极为频发的地质灾害,且其积累位移监测曲线有着复杂的非线性特性,对此各研究者建立过许多预测模型,然而这些模型的预测精度不尽如人意。基于Elman神经网络可以任意精度逼近任意非线性函数的特征,并以sigmoid为方程的核函数,在选择隐含层数时用了试用法,通过"3δ"法及归一化工程实例滑坡累积位移数据,建立了Elman神经网络动态预测模型。基于该模型对多个监测点数据进行动态预测,结果表明该模型的预测结果与实测数据的吻合度较高,且平均误差为1.78%,预测精度较高,验证了Elman神经网络能够在预测滑坡灾害中发挥一定作用。The landslide is a very frequent geological disaster in China, and its monitoring curve of the accumulative displacement has complex nonlinear property. Researchers have established many prediction models, however, the accuracy of these prediction models is not satisfactory. Based on the Elman neural network which can approximate any arbitrary nonlinear function by arbitrary precision, this paper takes the equation for sigmoid as the kernel function, and uses the method of trial when choosing hidden layer, and through the ＂3δ＂ method and normalized engineering instance of landslides to accumulate displacement data, and then Elman neural network dynamic prediction model is established. The model made dynamic prediction to the multiple monitoring data and the results show that the goodness of fit between the model prediction results and the measured data is quite high, and the average error is 1.78%, which means that the prediction accuracy is relatively high, which can verify the Elman neural network can play a role in the prediction of landslide disasters.

关 键 词：ELMAN神经网络 3δ法 动态预测 核函数 

分 类 号：P642.22[天文地球—工程地质学]