建筑物震陷预测新方法研究  

NEW METHOD OF STUDY ON PREDICTION OF BUILDING SETTLEMENTS DUE TO EARTHQUAKE LIQUEFACTION

作　　者：刘义建 刘勇建[2] 

机构地区：[1]湖南省地勘局407队,湖南怀化418000 [2]广东工业大学,广州510500

出　　处：《地质灾害与环境保护》2002年第3期47-50,共4页Journal of Geological Hazards and Environment Preservation

摘　　要：利用人工神经网络的基本原理 ,本文修正了经典 BP型神经网络的激励函数 ,并对学习率和训练样本进行了动态调整等多方面改进。根据 70个多层建筑震陷的实测资料 ,在分析了建筑物震陷的影响因素基础上 ,提取了 9个指标 ;采用改进后的 BP算法 ,建立了多指标的建筑物震陷预测模型。研究结果表明 ,改进的 BP网络性能良好 ,所建立的模型预测精度高 ,具有一定的工程实用价值 ;神经网络法是一种有效可行的预测新方法 。Based on the principle of artificial neural networks, the back-propagation (BP) algorithm is improved by modifying prompting function and a series of dynamic adjustment including learning ratio and training samples. According to the measured data from 70 multi-layer building settlements due to earthquake liquefaction, and by analyzing the effecting factors of building settlements, 9 indexes are distilled and a model is established for the prediction of building settlements due to earthquake liquefaction by adopting improved BP algorithm. The research results show that the improved BP networks has excellent performance, and that the predicting model works well and can satisfy the requirement of engineering in practice.

关 键 词：建筑物 震陷预测 BP网络 地基失效 地震 人工神经网络 

分 类 号：TU352.1[建筑科学—结构工程] P315.9[天文地球—地震学]