流量(数学)
环境科学
数据同化
流动条件
机械
同化(音韵学)
气象学
潜热
流动特性
流动阻力
工艺工程
作者
Huiting Chen,Wangyan Li,Jie Bao,Yansong Shen
标识
DOI:10.1016/j.cej.2026.179496
摘要
Gas-solid flow reactors are complex dynamical systems in chemical engineering, characterised by highly coupled, multi-scale physical fields that evolve over space and time. Accurate and efficient prediction of these reactors is essential for monitoring and optimisation, yet remains challenging due to limited and noisy measurements for validation, as well as the prohibitive cost of large-scale high-fidelity simulations. To overcome these limitations, this study proposes a correction-enhanced reduced-order modelling (ROM) approach with multi-scale latent assimilation for gas-solid flow reactors. The method conditionally corrects the prior predictions obtained from a data-driven ROM by assimilating observations in a low-dimensional latent space. In the offline stage, multi-scale Proper Orthogonal Decomposition and β -Variational Autoencoder (mPOD– β -VAE)compress the system into multi-scale latent spaces, significantly reducing both dimensionality and computational cost. During the online stage, forecast errors between the encoded observations and the ROM prediction are checked. When this error exceeds a dynamically determined threshold value, latent assimilation is triggered to update the prior latent state using an ensemble Kalman filter. The corrected latent state is subsequently decoded and reconstructed to obtain a more accurate state estimate. This methodology is demonstrated using a lab-scale fluidised bed reactor. The results show that the proposed framework achieves superior reconstruction of both observable and unobservable variables, reduces latent-state prediction errors through stable and well-conditioned assimilation updates, and delivers a substantial reduction of computational load. This approach reduces the computational complexity by 99.8% compared to conventional CFD-DEM and improves predictive accuracy by 44.4% over a standalone ROM, while maintaining a computational cost comparable to the standalone ROM. Overall, the proposed approach offers a promising tool for real-time prediction and monitoring in chemical engineering applications, paving the way towards a next-generation digital twin for gas-solid flow reactors.
科研通智能强力驱动
Strongly Powered by AbleSci AI