Deep Learning for Operation and Maintenance of Urban Lifeline Systems: A Comprehensive Review

Urban lifeline systems are critical infrastructures that ensure the stability and resilience of modern cities, yet they are increasingly stressed by aging assets, rapid urbanization, and climate change. Deep learning (DL), with its ability to automatically extract spatial, temporal, and topological features from multisource data, offers significant potential for advancing the operation and maintenance (O&M) of these systems. This study conducts a comprehensive review of DL applications across six major lifeline sectors, i.e., water, transportation, power, gas, district heating, and telecommunications. By integrating bibliometric and systematic analyses of 366 peer-reviewed articles, the review identifies five key O&M application scenarios: condition monitoring; load forecasting; intelligent scheduling; risk assessment; and emergency response. A structured status matrix is proposed to map DL architectures to these scenarios, highlighting maturity levels, sectoral imbalances, and underexplored opportunities. The study further discusses prevailing challenges and outlines future research directions for O&M of urban lifeline systems.