Machine learning and simulation models for predicting infrastructure systems' resilience

Our modern cities rely heavily on increasingly interdependent infrastructure systems to function smoothly. However, with climate change events on the rise, a single flood or heatwave may be able to cause multiple disruptions to interconnected power supply, water pump or transportation systems. How can infrastructure networks improve their resilience in the face of such unpredictability?

A multi-institutional team of researchers – including Dr Beatrice Cassottana and Dr Srijith Balakrishnan of FRS and CREATE colleagues from the external pageAdvanced Digital Sciences Center (ADSCcall_made) – have external pagepublished a new papercall_made for better resilience analysis. They merge Machine Learning (ML) algorithms with a simulation platform called InfraRisk, which simulates the performance of interdependent infrastructure networks.

By training ML models using InfraRisk-simulated data to predict system recovery and resilience, the team was able to examine key points of weakness, and to recommend strategies to strengthen the resilience of an interdependent infrastructure network.

For example, the team suggests decoupling the power and water systems on an infrastructure grid, necessitating back-up generators for water-pumps and strategic locating of such utilities to prevent cascading failures due to a disaster. These insights will help inform urban planning stakeholders – from infrastructure system operators to government infrastructure planners – to build a more resilient city in the face of unpredictable climate events.