基于神经网络的简支梁桥预应力衰减评估模型

以后张法室内模型梁受弯全过程试验采集的数据为样本,研究了预应力混凝土桥梁有效预应力与各主要影响因素之间复杂的非线性关系,应用改进BP神经网络建立了混凝土桥梁服役期内预应力衰减的神经网络评估模型。该模型以受弯梁全过程P-Δ理论曲线为依据,构造了基频衰减率(IFDETR)、变形指标(IDISP)、混凝土应变变化率(ICSV)、钢筋应变变化率(IRSV)4种归一化指标作为网络输入值,以计入时变效应的预应力衰减率指标(IEPDR)作为网络输出值。测试样本的评估结果表明,误差小于1%的测试样本数在样本集中所占的比例大于85%,且误差均小于单因素回归方法。

Evaluation model on prestress attenuation of simple spanbridges based on neural networks

The datum obtained by in-door experiments on post-tension concrete beam were applied to BP neural networks as the sample,the complicated non-linear relationship between effective prestress of PC structures and essential influencing factors was studied and the evaluation model of prestress attenuation was developed. Based on typical load-deflection curve of beam in bending process,such dimensionless coefficients as IFDETR(frequency deterioration ratio),IDISP(displacement index),ICSV(concrete strain variation index),IRSV(reinforcement strain variation index)and IEPDR(effective prestress deterioration ratio)were defined as input and output value, which made material and geometrical property independent.The analysis results show that the proportion of verification samples whose errors are within 5% is greater than 85%.Compared with the regressive evaluation method,the errors given by NN model are also much lower than that of it.3 tabs,5 figs,10 refs.