To reduce the risks of thermal runaway and fire in lithium-ion batteries under complex operating conditions, this paper proposes a machine learning-based operational risk assessment method for battery systems. Firstly, status of health (SOH), status of safety (SOS), and thermal gradient features are extracted at both the battery and system levels from operational monitoring data to construct a multi-dimensional risk indicator system. Secondly, based on multi-physics coupling characteristics, a safety operation risk evolution model is developed by integrating the effects of temperature fields, flow fields, and state parameters on the security of energy storage systems. Finally, a data-driven quantitative risk assessment model is established to correlate health status, safety status, and thermal gradients with safety grade classification. It provides technical support for the operational safety of lithium-ion battery systems, contributing to improved accuracy in risk prediction and enhanced capability for proactive prevention.