Adaptive multi-objective optimization for emergency evacuation at metro stations

• Main Authors: Guo, Kai, Zhang, Limao
• Other Authors: School of Civil and Environmental Engineering
• Format: Article
• Language: English
• Published: 2022
• Subjects: Engineering::Civil engineering; Emergency Evacuation; Metro Stations
• Online Access: https://hdl.handle.net/10356/162076
• Institution: Nanyang Technological University

Evacuation is a critical issue at metro stations, where damage, or even death, could be caused due to unexpected accidents if without proper evacuation. Multi-objectives are often desired in the evacuation management, and evacuation strategies should be tailored due to the dynamic features of overcrowded passengers and high uncertainty at the metro stations. A simulation-based approach integrating Light Gradient Boosting Machine (LightGBM) and Non-dominated Sorting Genetic Algorithm III (NSGA-III) is proposed to realize the automatic evacuation evaluation and adaptive optimization at metro stations. A framework consisting of 3 objectives and 9 influential factors is developed for the evacuation evaluation and adaptive optimization. A LightGBM based meta-model is used to construct the relationship between influential factors and objectives. A LightGBM and NSGA-III integrated optimization algorithm is employed to automatically evaluate the evacuation events and seek the adaptive strategies for the station renovation in order to achieve a safe evacuation under specific conditions. A station model simulating a metro station in Singapore is constructed to test the effectiveness and applicability of the proposed approach. It is found that (1) Among 50 target station based cases, 33 failing cases (at least one of the objectives fails) are identified, and the automatic evaluation results indicate the studied metro station could successfully evacuate the passengers when the passenger volume is no more than 1000, but it is very likely to fail when a higher volume exists; (2) Adaptive optimization strategies can be found for every different scenario, and an average 24.8% of the improvement degree can be achieved for all scenarios; (3) The adaptive multi-objective optimization is more cost-effective, presenting an average cost efficiency 2.04, which is significantly higher than the average of the non-adaptive optimization cost efficiency, 0.57. The novelty of this research lies in that (a) A LightGBM-based meta-model is built to construct the relationship between desired multi-objectives and the influential factors, which lays the foundation for the evacuation management from multi-objective optimization perspective; (b) The LightGBM and NSGA-III integrated model can realize the automatic evacuation evaluation and accordingly provide the adaptive strategies under specific conditions.