Machine learning improving seismic infrastructure recovery time estimation

• Published in: Bridge Maintenance, Safety, Management, Digitalization and Sustainability pp. 3276 - 3285
• Main Authors: Makhoul, N., Derras, B.
• Format: Book Chapter
• Language: English
• Published: CRC Press 2024
• Edition: 1
• ISBN: 1032775602; 9781032775609; 1032770406; 9781032770406
• DOI: 10.1201/9781003483755-388

To maintain infrastructure functionalities and ensure the rapid recovery of their functionalities, during powerful earthquakes, an assessment of infrastructure's seismic resilience is essential. One of the crucial elements of an investigation of seismic resilience is recovery time (RT). It is essential to make sure that the RT is anticipated in advance. This RT relies on various factors (Derras & Makhoul 2022). Unfortunately, the full set of factors that likely impacted RT are not always available. For this reason, we adopt in this study, an approach that is based on the simulation of the missing factors from those already available (Derras & Makhoul, 2023). To prove this, several machine learning algorithms have been adopted. This reveals that the "Boosted Trees Regression" algorithm is the best. Sensitivity analysis revealed that "system type" is the most important factor, followed by cumulative absolute velocity (CAV). The model thus developed is validated by dividing the data set into three "training-validation-test" subsets. The methodology, used in this study, allows us to prevent overfitting problems and to obtain an optimal aleatory variability of DT represented here by RSME (sigma). The best present model (Boosted Tree algorithm) allows us to reduce this variability at a rate equal to 51 %. It can offer specialists an approximate value of recovery time, an important piece of information when assessing seismic resilience.