煤炭超临界水制氢反应器多相流降阶建模及快速校正预测的研究

Study on Reduced Order Modeling and Calibrated Prediction of Multiphase Flows in Coal-Supercritical Water Fluidized Bed Reactor for Hydrogen Production

煤炭超临界水制氢技术将超临界水用作煤气化反应媒介,具有环境友好、煤炭转化效率高等优势。本文建立了煤炭超临界水制氢反应器二维瞬态数值模型,采用动力学模态分解(Dynamic Mode Decomposition,DMD)方法对各组分流场数据进行低维空间动力学降阶分析,针对传统DMD方法流场预测时的误差累积问题,提出基于最近邻和线性回归校正策略的DMD单步预测方法。结果表明:利用DMD方法可以较准确地提取制氢反应器内多相流各物理量的主导模态及其时变特征;与传统的DMD预测方法相比,基于修正策略的DMD单步预测方法在训练集上的重构精度相仿,但对未来时刻流场演变的预测精度可提高1~2个数量级;与基于最近邻校正策略的DMD单步预测相比,线性回归校正策略的预测精度更高,但重构精度略低,计算复杂度相对更高;另外,增加DMD截断阶数与训练数据量有利于提升模型重构与预测精度。

Coal-supercritical water hydrogen production technology uses supercritical water to achieve coal gasification reaction,which is environmental friendly and enjoys high conversion efficiency.In this paper,a two-dimensional transient numerical model of coal supercritical water hydrogen production reactor is established,and the Dynamic Mode Decomposition(DMD)is employed to analyze the fow fields.Aiming at the error accumulation problem of traditional DMD method in fow fields prediction,this paper proposes the nearest neighbor and linear regression error calibration strategies to achieve the DMD one-step-ahead prediction.Extensive comparative results have revealed that 1)the dominant modes and time-varying characteristics of various physical quantities of multiphase fows in the reactor can be obtained by the data-driven DMD method;2)the reconstruction accuracy of error-calibrated one-step-ahead prediction is similar to the traditional DMD prediction on the training set,but the prediction accuracy of flow fields evolution in the future period could be improved by 1~2 orders of magnitude;3)compared to the nearest neighbor strategy,the linear regression strategy has higher prediction accuracy,but it suffers from slightly worse reconstruction accuracy and higher computational budget;and finally;4)increasing the DMD truncation order and the number of training data could improve the reconstruction and prediction accuracy.