A long-term vertical displacement prediction method of concrete bridges based on meteorological shared data and optimized GRU model

Abstract

Based on data from the meteorological shared data platform, this study proposes a method for predicting the long-term vertical displacement (VD) of concrete bridges by integrating the Northern Goshawk Optimization (NGO) algorithm with the Gated Recurrent Unit (GRU) network. This method can be employed to safely assess concrete bridges and recover missing VD data. Specifically, it uses historical meteorological information and time information provided by the meteorological shared data platform (European Centre for Medium-Range Weather Forecasts) to generate the input parameters of the GRU model. It then employs the long-term VD data from the concrete bridge structural health monitoring system to produce the output parameters of the GRU model. Moreover, the hyperparameters for the GRU model training are optimized using the NGO algorithm. Four NGO-GRU models with different input conditions are proposed, taking the long-term VD prediction of a prestressed concrete bridge as a case study and considering the correlation between different meteorological factors and VD and the long-term time-dependent effects on concrete structures. Through a comparative analysis of the model’s prediction performance of multiple sensors under different conditions, it is found that the NGO-GRU model achieved the best prediction performance when using a combination of air temperature, time information, air pressure, solar radiation intensity, and wind speed as inputs, with a prediction error of less than 6.00%. Furthermore, compared with the benchmark models, the NGO-GRU model demonstrated the highest accuracy in VD prediction. Under the optimal input conditions, the prediction performance of the NGO-GRU model improved by 16.46% to 46.17% compared with the other models, validating the robustness and effectiveness of the proposed method.