Thesis of Miadana Valisoa

Defense of thesis Miadana Valisoa - Laboratory PC2A

Abstract :

Underground railway stations (URS) often exhibit concentrations of particulate matter (PM10 and PM2.5) that exceed the exposure limits recommended by the World Health Organization, as well as those measured in outdoor air. This situation can have negative impacts, both on the health of travelers and employees and on the image and attractiveness of rail transport. Aware of these issues, since 2016 SNCF has been conducting measurement campaigns to monitor PM2.5 and PM10 across its underground stations in the Île-de-France region. Among these stations, three have been continuously instrumented for several years. Moreover, air filtration experiments at platform levels have been carried out to improve air quality in these specific environments. However, analyzing and interpreting these measurements remains challenging, particularly in defining reference values. This is why SNCF wished to explore and develop new approaches.
The work presented in this thesis focuses on the development of new analysis methods, adapted to the specific dynamics of URS, especially the very strong fluctuations in particle concentrations. Using long-term measurements (several years), we have developed a rigorous and robust methodology to determine typical daily profiles in each station and a daily amplitude coefficient (DAC), distinguishing weekdays from weekends. Our results reveal a seasonal evolution of concentrations, with higher levels in summer and lower in winter, which differs from the trends reported in the literature in URS and outdoor air. The DAC approach allows us to disregard seasonal fluctuations to analyze the long-term evolution of particle concentrations. It also facilitates comparisons between different stations and contributes to identifying some parameters influencing particle concentrations, such as train frequency, the presence of ventilation systems, the type of station (underground or partially underground), and the railway line concerned.
A more detailed analysis was carried out using Generalized Additive Models (GAM). It helped identify the most significant factors and quantify their impact, including the time of day, reflecting train frequency, the period of the year, and CO2 levels. The effects of temperature and humidity, measured at the platform level, are less pronounced, and the contribution of outdoor air is very low, suggesting that the main factors influencing particle concentrations mainly originate from within the URS.
Finally, three depollution experiments were evaluated: a positive ionization technology, a wet technology, and traditional filtration. The results obtained via DAC and GAM show variable effectiveness depending on the technologies and particle classes (PM10 and PM2.5), with efficiency rates ranging from 0 to 34% depending on the technology used. We also implemented a model of recurrent neural networks (LSTM), which allows for analyzing the effectiveness of these filtration systems almost immediately and not a posteriori of the concentrations.
This thesis thus proposes a rigorous and adaptable methodology for understanding the dynamics of particle concentrations in URS, applicable to particles but also to other pollutants. The protocols developed provide immediately usable tools for SNCF, which can use them to sustainably improve air quality in these complex environments. Recommendations for future measurement campaigns have also been outlined.

Keywords : air quality,indoor pollution,particulate matter,underground railway facilities,filtration systems