A novel bridge wind-induced vibration response prediction algorithm based on temporal convolution network.

Area of Science:

• Structural Engineering
• Computational Mechanics
• Data Science

Background:

• Bridges with high piers and long cantilevers are vulnerable to wind-induced vibrations during construction, posing significant safety risks.
• The nonlinear and random nature of wind vibration responses presents challenges for accurate prediction during the construction phase.

Purpose of the Study:

• To propose a novel prediction algorithm for bridge wind-induced vibration response during construction using a temporal convolutional network (TCN).
• To evaluate the prediction accuracy and generalization ability of the TCN-based model against other advanced algorithms.

Main Methods:

• Utilizing causal convolution within the TCN to identify mapping relationships in wind-induced vibration acceleration data.
• Employing dilation convolution to capture multi-scale features of wind vibration response.
• Implementing residual connections to mitigate gradient vanishing issues in the network.

Main Results:

• The TCN-based model demonstrated superior prediction accuracy and generalization capabilities compared to Recurrent Neural Network (RNN), Long-Short-Term Memory (LSTM), and Gated Unit Network (GRU) models.
• The algorithm effectively predicted wind vibration acceleration in multiple directions, including torsion, vertical, transverse, and along the bridge.

Conclusions:

• The proposed TCN-based algorithm offers a robust and accurate solution for predicting wind-induced vibration responses in bridges during construction.
• This advancement contributes to improved safety and risk management in bridge engineering projects, particularly in high-wind environments.