Enhancing the sustainable management of fine particulate matter-related health risks at subway stations through sequential forecast and gated probabilistic transformer

Abstract

The alarming rise in passenger traffic within underground subway stations has prompted concerns about indoor air quality in these enclosed spaces. Implementing early warning systems, particularly those focused on fine particulate matter (PM2.5), is essential for maintaining a sustainable underground environment. Despite efforts to quantify air quality dynamics in these environments, the inherent uncertainties, and complex non-linear behavior of PM2.5 remain a significant challenge. This study combines a gated probabilistic transformer (PT-Trans) framework with a genetic algorithm (GA) based quantile scheduling approach. This methodology aims to overcome the constraints of deterministic forecasting methods, thereby enhancing PM2.5 early warning systems and ventilation control in underground environments. Performance evaluation revealed that, in comparison with conventional probabilistic networks, the PT-Trans yielded a sharp prediction interval, leading to a reduction in Winkler score up to 34.4%, with a coverage probability of 87.1%. The GA-enabled PT-Trans (GA-PT-Trans) increases the critical success index for 5-h ahead sequential health risk assessment by 12.4% compared to the deterministic model while decreasing the missed alarm rate for moderate health risk levels by 72.5%. The ventilation based on GA-PT-Trans prediction also improved and reduced PM2.5 concentrations by 19.7%. The proposed framework offers a significant contribution towards the precise analysis and mitigation of health risks associated with underground building environments. This advancement is particularly pertinent today, as it addresses a critical barrier to the integration of automated monitoring and control within the emerging field of smart buildings.