关键词: 桥梁工程；重载车辆；深度学习；信号处理；抗噪性分析

Abstract: In order to ensure the safety of the bridge in the environment of heavy-load vehicles, the bridge response was predicted. Firstly, a simply supported box girder and two-axle heavy-load vehicles were selected as the research objects, Abaqus software was used to establish the bridge model, and the dload subroutine was used to simulate the driving process of heavy-load vehicles. Secondly, the complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) signal decomposition model was established to improve the response prediction effect. Then, a long short term memory (LSTM) cyclic network was constructed to train and predict the decomposed signal, and compared with the single LSTM model. Finally, in order to verify the prediction ability of the proposed model under actual noise data, the data complexity was increased by adding a specified signal-to-noise ratio (SNR) noise data, and then the stability of the proposed model was verified. The results show that applying the CEEMDAN decomposition algorithm to the LSTM model for predicting the acceleration, deflection and strain of bridges under heavy-load environment can effectively improve the prediction effect of the proposed model, with improvements in all indicators.

Key words: bridge engineering; heavy-load vehicles; deep learning; signal processing; noise immunity analysis