基于改进Elman神经网络抽水地面沉降预报模型  

Prediction Model of Land Subsidence Caused by Groundwater Pumping Based on Improved Elman Neural Network

作　　者：赵宝民[1] 于德浩[1,2] 

机构地区：[1]沈阳军区司令部工程科研设计所 [2]65056部队

出　　处：《污染防治技术》2010年第5期1-3,共3页Pollution Control Technology

摘　　要：为了提高普通Elman神经网络模型在预测抽水地面沉降过程中稳定性差、预报精度低的问题,提出了一种改进Elman神经网络模型。该模型在结构单元中,增加一个固定增益α的自反馈连接,从而可以消除普通Elman网络对结构单元连接权的学习稳定性差、以致不能提供可接受的逼近精度的问题。从预报结果来看,改进Elman网络的预测值与实测值拟和度更高,误差更小,预报精度比普通Elman网络提高了5-23倍;能够很好地反映抽水当天地面沉降的真实情况。证明了改进后的Elman网络模型具有更高的可靠性和实用性。Aimed at solving the problems of Elman neural network model used in land subsidence, such as low stability and prediction, an improved Elman neural network model was presented in this paper. A fixed - gain a self - feedback connection was added to construction unit, which could solve the above problems. According to the results, the precision of improved Elman neural networkmodel is from five times to twenty - three times as high as that of Elman neural network. The predicting results of improved Elman neural network can represent the real land subsidence in that day, which testifies the model having higher reliability and practicalbility.

关 键 词：改进ELMAN神经网络 学习算法 抽水地面沉降 预报模型 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]