【摘要】 在传统的BP神经网络中引入学习速率自适应调整算法,通过多次数值模拟计算确定学习速率和动量系数等网络关键参数的取值;分析了学习速率、动量系数等网络参数对网络收敛速度和输出精度的影响;探讨了训练样本容量与网络识别效果的关系.分别使用挠度、挠度曲率、应变和应变曲率作为输入参数对桁架桥梁荷载进行识别.结果显示以挠度曲率或应变曲率为输入参数的网络识别效果明显优于以挠度或应变为输入参数的网络,以应变为输入的网络识别效果优于挠度的情况;学习速率自适应调整算法有效避免了网络训练过程中误差曲线振荡现象的产生,提高了网络的学习效率,网络关键参数的最优取值改善了网络的收敛速度和输出精度.更多还原

【Abstract】 The self-adaptive algorithm of learning rate for BP neural networks was implemented to identify loads exerted on bridge structures.Values of the network’s key parameters such as initial learning rate and momentum factor were optimized by numerical simulation for several times.Training specimen capacity was varied to obtain different output precision,and the relationship between the specimen capacity and the identification effect was discussed.The displacement(or displacement curvature) and strain(or strain curvature) were used as input data of the BP networks respectively.The identification effect of network using displacement curvature or strain as input data; networks using strain(strain curvature) as input data performed better than networks using displacement(displacement curvature)as input data.The developed algorithm can avoid oscillation phenomenon during the training process and improve learning rate effectively. Optimum values of network’s key parameters enhance error convergence velocity and output precision of the BP networks.更多还原