Analysis on Network Model Parameters of BP Neural Network in the Assessment for Bridge Reliability

Abstract

Using of back propagation (BP) neural network model to evaluate the bridge reliability would not only overcome the shortcomings of the traditional assessment methods, but also present many features such as dynamic adjustment, high precision, high efficiency and strong portability. Through analysis experiments on the parameters influenced by the learning samples, the learning rates, the hidden layer nodes and the initial weights in the reliability assessment of Masangxi Yangtze River Bridge, Proposed a 14-16-1 BP neural network with the learning rate of 0.005 to evaluate the reliability of Masangxi Yangtze River Bridge, which has 1000 groups of learning samples. This model has higher precision and assessment efficiency, which make some useful exploration for this intelligent algorithm model applied in other bridges’ reliability assessment.