Evolutionary identification in dense separation fluidized beds using dynamic mode decomposition with pruning

Abstract

Evolutionary identification of hydrodynamics from pressure signals is crucial for advancing the precise control of dry coal separation. Dynamic Mode Decomposition (DMD) is the key method to construct the data-driven control framework. Pressure signals rather than snapshots are investigated for industrial applications, bringing challenges to the implementation of DMD. The techniques of time delay embedding and optimal amplitude are introduced to make DMD work better for pressure signals. Comprehensive parameter tests of stack dimension $s$$s$ and truncation order $r$$r$ are carried out to seek for optimal identification performance. Due to parameter sensitivity, the qualification verification by sliding windows is performed to determine the robustness of parameter pairs. Spatiotemporal coherent structures are extracted to guide the regulation of separation process. In order to avoid the inefficiency of control, a heuristic sparsity promoting method using pruning is proposed to obtain a reduced order model. The original modes more than 100 can be reduced to approximately 35 primary modes. Furthermore, the Prune dominant frequency is defined, which can perceive the subtle fluctuations of temporal evolution than FFT and DMD for the long-term time. Present study provides the insight of hydrodynamics of dense gas-solid fluidized bed, establishing the foundation for future control studies of dry coal separation.