关于进化神经网络混凝土大坝变形预测_李守巨.pdfVIP

第24 卷第4 期 岩 土 力 学 Vol.24 No.4 2003 年8 月 Rock and Soil Mechanics Aug. 2003 文章编号：1000－7598 －(2003) 04―0634―05 基于进化神经网络混凝土大坝变形预测 1 1 2 李守巨 ,刘迎曦 ,刘玉静 (1.大连理工大学工业装备结构分析国家重点实验室,大连 116024； 2.长春工业大学基础科学系,长春 130012) 摘 要：根据丰满大坝多年变形观测数据，建立了基于进化神经网络混凝土大坝变形预测方法。经典的BP 神经网络的缺陷 在于收敛速度慢和泛化能力弱等特性。与普通的多元回归方法和传统的BP 神经网络相比，采用遗传算法训练的人工神经网 络预测模型预报大坝的变形具有精度高和全局收敛的特点。在丰满大坝工程实际应用表明，所建立的基于进化神经网络混凝 土大坝变形预报方法与广泛采用的统计方法相比，可以显著提高大坝变形预报精度。 关 键 词：人工神经网络；变形预报；混凝土大坝；遗传算法 中图分类号：TU454, O233 文献标识码:A Dam deformation forecasting by evolving artificial neural network 1 1 2 LI Shou-ju , LIU Ying-xi , LIU Yu-jing ( 1.State Key Laboratory of Structural Analysis of Industrial Equipment, Dalian University of Technology, Dalian 116023, China; 2. Basic Science Department, Changchun University of Technology, Changchun 130012, China) Abstract: Based on the measured data of Fengman dam deformations for many years, the artificial neural network evolved by a genetic algorithm was adopted for forecasting the dam deformation. The shortcomings of the traditional BP artificial neural network lie in the slowness in the convergence rate and the weakness in the generalization ability. Compared with the popular multi-factor regression model and BP artificial neural network, the forecasting model based on artificial neural network evolved by a genetic algorithm had the characteristic s of accurately forecasting and global convergence. The practical application to Fengman concrete dam shows that, compared with a commonly used statistical method, the forecasting method proposed can obv