A Legendre Neural Network-Based Approach to Multiparameter Identification of Traffic Loads Across the Full Spatiotemporal Domain

Abstract

The accurate identification of traffic loads acting on bridges provides an effective basis for the traffic control and operation of in-service bridges. To improve the efficiency and accuracy of loading identification, we propose an efficient multiparameter identification method with a Legendre neural network (LNN) for the monitoring of traffic loads across the full spatiotemporal domain. Compared to conventional studies that suffer from ill-posed problems and neural network-based means that lack a physically interpretable model, with the proposed strategy, both the explicit expression and time histories of the traffic load can be simultaneously obtained. Meanwhile, inaccurate load identification at the bridge’s supports, which is caused by ill-posed problems, does not exist in the identification process using the LNN. After the training and optimization of the LNN, its identification accuracy for speed and the magnitude of forces reached 98.6% and 98.3%, respectively. The results suggest that an identification method with a well-trained LNN is insensitive to noise.