SiGAN: A 3D sensor importance deep generative model for urban wind flow field monitoring

Urban wind fields play a pivotal role in shaping urban climate, ensuring the safety of drone operations and wind sensitive structures. The monitoring of urban wind fields can offer assurance in addressing these issues. Given the three-dimensional (3D) complexity of wind fields and the significant vertical variation in urban building layouts, a 3D approach to sensor placement is imperative for accurately capturing the wind environment. However, sensor deployment incurs substantial costs, and the presence of numerous irregular obstacles in 3D urban spaces complicates the process. Consequently, determining the optimal distribution of sensors to acquire more effective wind field information in such a complex environment has emerged as a significant research challenge. This study proposes a 3D urban sensor placement scheme that utilizes a custom-developed deep learning model called Sensor Importance Generative Adversarial Network (SiGAN). The effectiveness of the proposed scheme is evaluated using computational fluid dynamics (CFD) data from urban wind fields at three different heights. The results demonstrate that the proposed scheme effectively captures characteristic information of wind fields at varying heights and enables the arrangement of sensors from a global perspective. The sensor optimization scheme proposed in this study can reduce the wind field reconstruction error by approximately 10% compared to other schemes. Consequently, the proposed 3D sensor placement scheme allows for the careful design of sensor placement positions to address a range of wind engineering problems. This approach establishes a solid foundation for monitoring the 3D urban wind environment and resolving related downstream issues.