Adaptive optimization random forest for pressure prediction in industrial gas-solid fluidized beds

The establishment of a pressure prediction model in the gas-solid fluidization process enables acceptable forecasts of pressure drop, facilitating precise control and optimization of fluidized operations. Coupling effects between operational parameters and limited real-world samples further complicate model development. To address these issues, this paper proposes an industrial gas-solid fluidized bed axial pressure prediction model based on Scale-Invariant Feature Transform (SIFT) and Adaptive Optimization Random Forest (AO-RF). The SIFT module employs fixed distribution for data mapping, addressing the challenge of mismatch between feature and prediction data. Meanwhile, AO-RF effectively handles the complex relationships between limited samples and multi-scale data through Bayesian automatic hyperparameter optimization and robust model ensemble techniques, accurately capturing the nonlinear and dynamic characteristics of the fluidization process. Experimental results confirm the high prediction accuracy and generalization performance of the model, laying a solid foundation for AI (Artificial Intelligence) applications in industrial gas-solid fluidization processes. [Display omitted] •A gas-solid fluidized bed axial pressure prediction model was proposed based on machine learning method.•Scale-Invariant Feature Transform was used to address predictive performance loss from data mismatch across scales.•A multi-label prediction model is built with Random Forest to capture the nonlinear and dynamic aspects of fluidization.•Bayesian Optimization algorithm was adopted to adjust parameters automatically, improving the accuracy of model.