Predictive Modeling for Safety Risk Assessment in Bridge Construction using ARIMA and LSTM

Due to the heightened hazards associated with construction compared to operational phases, particularly in coastal regions, it is imperative to conduct safety risk assessments throughout the bridge construction process, especially for large and complex structures. The existing evaluation methodologies are inadequate for addressing risk correlations, as they rely on subjective weight assignments and do not reveal essential risk factors necessary for management and prevention. This study establishes the importance of safety risk indicators by employing deep learning algorithms on data received from many origins and categorized into two decision-making groups. A hybrid ARIMA-LSTM model is proposed for the effective management of nonlinear risk components. Upon estimating parameters and eliminating linear trends from the speed series data, the ARIMA model transmits the residuals to the LSTM network, which processes the normalised nonlinear components. Upon synthesizing both outcomes, we obtain our conclusive risk predictions. The model surpassed prior methodologies with an RMSE of 18.453, MAE of 15.287, and MAPE of 0.132, evaluated across two test driving cycles and two real-world driving scenarios. The enhanced safety management, augmented predictive accuracy, and support in decision-making for complex bridge projects illustrate the effectiveness of hybrid models in Safety Risk Assessment for Bridge Construction.