Abstract
This paper presents an intelligent Ventilation-on-Demand (VOD) control framework integrated with a Digital Twin Model of a Subterranean Ventilation System (DTM-SVS) to optimise airflow management for enhanced safety, energy efficiency, and operational adaptability in underground environments. DTM-SVS models the ventilation network using graph theory, with nodes representing airflow intersections and edges linked tunnels. Ventilation resistance is obtained using Ventilation Resistance Coefficient Estimation Models (VRCEM) to incorporate nonlinear behaviour governed by conservation of mass, energy, and fundamental ventilation laws. The intelligent VOD layer utilizes deep reinforcement learning (DRL) alongside real-time and historical ventilation data to estimate the Ventilation Resistance Coefficient (VRC) of different zones and tunnels. And it ensures the trajectory path of airflow aligns with operational demand and environmental conditions in the DTM-SVS. An experimental lab-scale ventilation platform is utilized to calculate VRC using operational demand and environmental conditions of zones. As a result, fan service life is prolonged and can saves 50-65% energy compared to constant-speed fan operation. The proposed approach is applicable for underground metro systems, tunnel networks, bunkers, and mines, enabling real-time, data-driven ventilation control for improved operational safety and sustainability.