Neighborhood preservation-based trajectory clustering for analyzing temporal behavior of dynamic systems
Bálint Levente Tarcsay, János Abonyi, Sándor Németh · Digital Chemical Engineering · 2025
This work presents clustering algorithms for identifying movement patterns in trajectories, with a focus on applications in chemical engineering. The exponential growth of dynamic system data necessitates algorithms that account for both local and global trajectory trends. Existing methods often overlook these aspects. We propose two DBSCAN-based variants that cluster trajectories from dynamic systems using agglomeration criteria reflecting the temporal evolution of object neighborhoods in phase space. The first algorithm groups objects with similar movement patterns over a defined observation period, while the second clusters objects with consistent neighborhood similarity over extended periods. These approaches enable the identification of localized neighborhood preservation and trajectory similarity, alongside global trends. We demonstrate the method by clustering particle trajectories generated via computational fluid dynamics, revealing characteristic flow regions within a tank equipped with static mixers. This highlights the methods’ utility for analyzing and optimizing dynamic processes in chemical engineering.