Forecasting and early warning of bridge monitoring information based on a multivariate time series ARDL model

Abstract

For bridge monitoring data, employing time series analysis to establish mathematical models that can reflect the operational state of a structure and predict its future state is a feasible research approach for bridge condition assessment. Existing studies primarily consider data from a single monitoring point. However, the bridge health monitoring system (HMS) is equipped with various sensors, and the monitoring data obtained from different measuring points are not isolated. Therefore, this study aims to utilize a multi-point, multi-variable perspective. The multivariate ARDL model, which utilizes multiple monitoring time series, is developed and the interrelationships between different measuring points and variables are investigated. Moreover, a method for structural forecasting and early warning that utilizes the ARDL model is developed. For the warning threshold interval, the confidence interval of the predicted values is utilized, and the establishment and dynamic updating of the periodic model is discussed. Finally, based on the health monitoring data of Kunshan Yufeng Bridge, the ARDL model for bearing displacement and temperature monitoring data is established. Using the model, this study explores the long-term and short-term quantitative relationships and the interactions between bearing displacement and temperature variables. Subsequently, forecasting and early-warning analyses are conducted. Compared with the ARIMA model, which is based on a single displacement variable, the forecasting MAE, RMSE, MAPE, and TIC of the ARIMA model are 2.38681, 3.34296, 0.08698, and 0.06215, while these metrics of the ARDL model are only 0.53197, 0.62786, 0.01913, and 0.01138. Apparently, the ARDL model exhibits much higher forecasting accuracy; thus, early warning precision is ensured. The early-warning and assessment method established herein is robust against bridge types or monitoring parameters; thus, it exhibits satisfactory applicability and feasibility.