摘要
BP神经网络已被广泛应用于多个研究领域,为克服和改善传统的BP算法收敛速度慢、易陷入局部极小、泛化能力差的缺陷,基于混凝土适筋梁挠度预测的实际工程问题,提出L-M算法,并对现场的原始数据进行分组取样;实验数据对比及算法仿真结果表明,基于改进的L-M算法的神经网络推算钢筋混凝土适筋梁长期挠度的方法是可行的,并且L-M算法比传统BP算法收敛速度快,挠度的预测结果比规范计算结果更加接近实测数据;应用结果表明,L-M算法利用二阶导数的特性,加快了收敛速度,提高了计算精度;而运用分组取样法则提高了网络的泛化能力。
The BP neural network have been widely applied in many research field,to overcome and improve the defect of traditional BP algorithm which has slow convergence speed,easily falls into relative minimum and bad generalization ability,based on the concrete poor defects of steel beam deflection forecast fitness actual project,a L-M algorithm is proposed,and divide the original data into groups to take samples.Experimental data contrast and algorithm simulation results show it is feasible to calculate reinforced concrete beam of long-term deflection based on improved L-M neural network optimal method,and the convergence speed of L-M algorithm is much faster than traditional BP algorithm's,the results of predicting deflection is closer to measurement data than the standard calculated results.The application shows that L-M algorithm uses the characteristics of second order derivative,accelerate the convergence rate and improve the calculation accuracy.And the use of grouping sampling method improved the network generalization ability.
出处
《计算机测量与控制》
CSCD
北大核心
2012年第1期50-52,58,共4页
Computer Measurement &Control
关键词
L-M算法
神经网络
混凝土适筋梁
挠度
预测
L-M algorithm
neural network
reinforced concrete beam fitness
deflection
predict