Prediction of temperature field indicators of concrete bridges based on the Informer model

Accurate assessment of bridge temperature distributions is essential for evaluating structural safety. To predict bridge temperature field indicators, this study proposes an Informer-based forecasting method. Structural temperature field data are generated through finite element thermal analysis, and the two-dimensional temperature field at a representative cross section is decomposed into four physically meaningful indicators. These indicators are combined with time and meteorological features to construct a sample database for model development and evaluation. The results show that, compared with the typical time-series forecasting models LSTM and ARIMA, the Informer model achieves higher accuracy in single-step prediction. As the forecast horizon increases, the prediction accuracy of Informer decreases to some extent but remains relatively high at the 24-h forecast horizon. Based on these evaluation results, a rolling prediction method with a 24-h cycle is developed, enabling continuous forecasting of bridge temperature field indicators.