Data-Driven Reduced-Order Model for Bubbling Fluidized Beds
Xiaofei Li, Shuai Wang, Dali Kong, Kun Luo, Jianren Fan
Abstract
Simulation of dense gas–solid flow in fluidized beds is a computationally intensive procedure, and emerging speedup simulation methods are still unsatisfactory. This work developed a pioneering data-driven reduced-order model (ROM) for efficient modeling of dense gas–solid flow in bubbling fluidized beds (BFB) by integrating the proper orthogonal decomposition (POD) and the radial basis function neural network (RBFNN). Specifically, this study extracts the fundamental eigenvectors of the gas–solid flow process and constructs a prediction function for the corresponding eigenvector coefficients. The effectiveness of this ROM is conclusively assessed by comparing it with the full-order model (FOM) in terms of simulated results and performance criteria. The results indicate that the 10-bases-ROM and 64-bases-ROM exhibit 50 and 90% of the energy, respectively, and achieve flow field reconstruction accuracy of 50 and 90%. Moreover, compared to the FOM, the 10-bases-ROM and the 64-bases-ROM demonstrate 700-fold and 120-fold increases in simulation efficiency, respectively. These findings indicate that the proposed model has the potential to be an effective tool for industrial engineering process predictions in real time.