基于Elman型神经网络的金川二矿地表岩移时序预测模型  

ELMAN NEURAL NETWORK BASED TIME-SERIES FORECASTING MODEL FOR GROUND SURFACE MOVEMENT ON NO.2 NICKEL MINE AREA IN JINCHUAN

作　　者：袁仁茂[1] 马凤山[1] 邓清海[1] 徐锡伟[2] 

机构地区：[1]中国科学院地质与地球物理研究所,北京100029 [2]中国地震局地质研究所,北京100029

出　　处：《工程地质学报》2008年第1期116-123,共8页Journal of Engineering Geology

基　　金：中国科学院知识创新工程重要方向性项目(KZCX2-YW-113);国家自然科学基金项目(40702048)资助

摘　　要：建立于煤矿开采基础之上的矿山开采沉陷理论和预测方法并不适用于象金川这样厚大、陡倾的金属矿床开采的岩移问题,因此,本文探讨利用神经网络来对地表岩移进行预测。根据Elman神经网络能够逼近任意非线性函数的特点和具有反映系统动态特性的能力,提出了利用Elman神经网络建立地表岩移时序预报模型的方法。利用金川二矿区GPS监测所得到的时间序列数据,通过对Elman神经网络模型预测值与GPS实测值之间的比较,结果表明模型预测显示了良好的准确性,特别是在时间步长较短情况下,应用于实际预测一定程度上可以弥补金属矿山岩移预测方法不足的缺憾。Artificial neural networks （ ANNs ） can be used for the ground surface movement prediction in the cases that traditional theories of subsidence and forecasting methods are not suitable, because they are based on the nonmetal mine underground mining. New methods are needed to deal with the ground surface movement problems in metal mine area such as Jinchuan Nickel mine with high dip angle. It is known that the Elman neural network can well approach any nonlinear continuous function and has ability to reflect dynamic features of the systems. Therefore, a time - series forecasting model of ground surface movement based on Elman neural network is presented. The datum of ground surface deformation got form GPS monitoring in Jinchuan Mine area were used to verify this model. Through comparing the forecasting result from the Elman model with the monitoring datum from GPS, it shows that the ANN prediction model is a useful method with good precision, especially under short time step prediction. The proposed method can offer a solution to the shortage of method in practice to a certain extent.

关 键 词：ELMAN神经网络 时序预测模型 地表岩移 金川镍矿 

分 类 号：TD824[矿业工程—煤矿开采]