PREDICTION OF BRIDGE DETERIORATION TREND BASED ON MACHINE LEARNING CONSIDERING CLIMATE CHANGE

Abstract

 The machine learning-based prediction models for bridge deterioration due to segmentation of deck, alkali-silica reaction (ASR) and frost damage are developed using environmental and meteorological data of bridges located in Tohoku region. LightGBM algorithm is applied to develop the models for solving the binary classification problem of deterioration/non-deterioration classes. The Shapley value is used to measure the contributions of input features to the output of the LightGBM prediction models for segmentation of deck, ASR and frost damage of bridges. The deterioration prediction models are validated based on the consistency between the deterioration factors determined by the Shapley values and the deterioration mechanism. Then, considering the changes in meteorological conditions under various climate change scenarios, the future deterioration trend of bridges located in Tohoku region is predicted using the machine learning-based prediction models. The prediction results with and without considering the effects of climate change demonstrate that the number of deteriorating bridges in the future depends on the climate change scenario. In addition, it is found that the bridges which would be deteriorated in the future are likely to be distributed around the bridges which are already deteriorated.