Optimizing block morphology for reducing traffic pollutant concentration in adjacent external spaces of street canyons: A machine learning approach

Abstract

Residential blocks located in near-road environments are exposed to traffic-related air pollution that adversely affects the health of residents. Pedestrian height areas in adjacent external spaces (AES) of street canyons are most severely affected by traffic pollutants. In order to find block morphological optimization strategies that can mitigate the concentration of pollutants in the area, machine learning (ML) regression models were investigated to replace the time-intensive computer fluid dynamic (CFD) model. Data from 1000 CFD simulation cases were used to train and test the ML models. The best surrogate model was selected and coupled with optimization algorithms. Both single-objective and multi-objective optimization problems based on regional concentration parameters were discussed. The results showed that the block's standard deviation of building height (BHstd) was an essential morphological parameter that can effectively impact the pollutant concentration. The effect mechanism of the building height at different locations in the block on the concentration of the pollutant in the different regions also varies. The optimum block configuration obtained from the optimizations coupled ML model can facilitate reducing pollutant concentrations. This study provides a framework for residential block design optimization that can help urban designers efficiently evaluate and select block layouts conducive to reducing the impact of traffic pollutants.