An AI-assisted solution for low-cost high-resolution urban scale environmental simulation
Funding agency
Fei Chi En Education and Research Fund, 2023-2024
Project introduction
Numerical simulation is an important approach to investigate urban environmental issues for low carbon city and sustainable development. However, the application of urban scale simulation for micro-climate management is restricted by its computational costs and spatial/temporal resolutions. To address this research gap, we propose an innovative simulation framework as a practical solution to manage the trade-off between computational cost and simulation resolution for urban scale climate studies. The framework couples mesoscale simulations performed by an existing mesoscale meteorological model and microscale simulations conducted by a newly-developed artificial intelligence model. It decomposes the large urban area into multiple small sub-domains and adopts the artificial intelligence model to simulate the microscale climate of each sub-domain and finally combines the sub-domain results to reconstruct the urban climate information of the large area.
The proposed research has four major research tasks. First, we plan to develop an automatic tool for converting geographical information system (GIS) data to computational fluid dynamics (CFD) geometry data to accelerate the procedure of CFD modelling in generating the database that is required to train the artificial intelligence model. Second, we will create a high-quality database for training and tuning the artificial intelligence powered surrogate model. Urban morphology clustering and feature-driven algorithm will be developed to minimize the data redundancy and ensure the feature diversity. Then, an artificial intelligence model will be developed for high-resolution microscale urban climate simulation, which is the core task in this research. Finally, the proposed simulation framework will be validated using field measurement data. The proposed research shifts the paradigm for large-scale urban climate simulation which is previously computationally expensive if not totally impossible.
