Deep Learning-Based Cable-Stayed Bridge Force Safety Warning Research

Among large-span bridges,cable-stayed bridges are one of the most common types of bridge.It is worth mentioning that China currently has the largest number of cable-stayed bridges in the world.Cable-stayed bridges are subjected to a combination of vehicle loads,wind loads,fatigue,seismic loads,human factors as well as material ageing and cracking over a long period of time,which can easily lead to structural failure problems and can pose a huge safety hazard.The main component of a cable-stayed bridge is the cable-stayed wire,which provides vertical tension.The health of a cable-stayed bridge can be affected by large continuous changes in cable force,so effective prediction of cable force is essential to improve the safety and stability of a cable-stayed bridge.The time series of cable force has certain periodic and non-smooth characteristics,so it is difficult to predict it accurately by a single prediction model.Based on this,this thesis adopts a deep learning method combining feature fusion and cable force signal decomposition to predict the cable force.This thesis firstly addresses the problem that long-term dependencies cannot be captured in long-term time series prediction and important information is easily ignored,and proposes a two-way long-and short-term memory network combined with convolutional neural network based on the attention mechanism for forecasting the solex.The model converts the original solex series into a smoothed solex time series by exponential sliding average method,inputs the information of the hidden layer of the temporal unit into the bidirectional long and short-term memory network model,independently assigns feature weights to the forward and backward long and short-term memory networks through an improved attention mechanism to enhance the feature extraction effect of the bidirectional long and short-term memory network,extracts local features through the convolutional neural network,and finally sends the data to the hidden layer of the bidirectional long and short-term memory network.The final data is sent to the hidden layer of the two-way long-and short-term memory network,and the output of the solex prediction results.Secondly,to address the problem of modal confounding when dealing with non-stationary time series with fluctuating characteristics,a two-way LTMN combined with convolutional neural network based on ensemble empirical modal decomposition is proposed.The model reduces the effect of non-smoothness of the force data by decomposing the force data into several non-interfering subseries of intrinsic modal functions through an ensemble empirical modal decomposition algorithm.The global features and local features of the time series are extracted using a two-way long and short-term memory network and a convolutional neural network,respectively,and then fused into the combined force prediction model to obtain the prediction results for each sub-series.Finally,the prediction results are obtained by superimposing the prediction results of all sub-series.Finally,to verify the prediction effect of the model proposed in this thesis,the data of a cable-stayed bridge in Xiamen from 2019 to 2020 were selected for the experiment,and the root mean square error,mean absolute error,mean absolute percentage error and fit score were used to evaluate the experimental results.The experimental results show that the model proposed in this thesis outperforms other methods in all evaluation indexes,and the model proposed in this thesis can predict the trend of cable force change more accurately,and use the isolated forest algorithm for anomaly detection to provide effective prevention and control measures for the prevention of sudden changes in cable force.