Data-driven optimisation of the ventilation systems in underground car parks: A CFD-DNN study

In this study, Autonomous Characterisation and Calibration via Evolutionary Simulation (ACCES) technique is integrated with computational fluid dynamics (CFD) to develop a systematic approach for configuring underground car park ventilation systems. The analysis seeks an optimum configuration applicable to both pollution control and smoke clearance phases. The objective functions are to achieve a specific velocity in the car park at a height of 1.7 m and to minimise the fans' flow rates. The ACCES algorithm then dynamically adjusts the fans' locations and optimises their flow rates. The aim is to reduce stagnant areas in the pollution control phase efficiently, ensuring that the minimum necessary flow rate is used to achieve effective pollution control in a cost-effective manner. The results revealed an optimum configuration for which only 1.67 % of the car park volume was stagnant. Additionally, the developed algorithm was found to be independent to the initial guess for fans locations/flowrates, where even defined in a profoundly biased manner they still converged to values quite neighbouring to that of the optimum case, with maximum relative difference (compared to optimum case) of 11 %. This suggests that the optimum configuration proposed by the ACCES may be the global optimum. This optimum configuration is also tested for smoke clearance phase, where the relevant fire safety standards were acceptably satisfied. Notably, the proposed methodology is generalisable and can be applied to any geometry with any constraints.