Randomized Time-Delayed Dynamic Mode Decomposition Approach for Reconstruction of Wind Pressure Fields on High-Rise Buildings

Dynamic mode decomposition (DMD) provides an effective approach for analyzing complex wind pressure fields by decomposing them into spatiotemporal modes. However, traditional DMD suffers from two main limitations: low spatial resolution caused by sparse pressure tap arrangements in wind tunnel tests and high computational costs when using time-delay embedding. To overcome these issues, this paper introduces randomized time-delayed DMD (RTD-DMD), a DMD algorithm that minimizes memory usage and computational cost while preserving accurate identification of coherent flow structures. The method employs randomized QB decomposition (an approximation algorithm for the QR decomposition) to compress the high-order Hankel matrix produced by time-delay embedding, followed by DMD for modal analysis of the reduced data set. Tests conducted on fabricated patterns data sets with artificial noise and real wind tunnel measurements confirm that RTD-DMD surpasses conventional DMD in both computational speed and reconstruction accuracy. By balancing dimensionality reduction and computational efficiency, RTD-DMD offers a scalable solution for processing large-scale, noisy wind pressure data, making it suitable for long-term monitoring and advanced aerodynamic flow studies in structural engineering applications.