To address the data scarcity challenge in microreactor-based polyurethane synthesis—caused by complex reaction mechanisms and high experimental costs—we propose a general few-shot modeling strategy based on the Transformer architecture,termed TabPFN.Unlike conventional transfer learning approaches that rely on source-domain data,this model leverages its pre-trained general inference capability to build a highly accurate predictive model using only 67 experimental samples (achieving R2>0.95 and MAPE<3.45% on the test set).Comparative experiments demonstrate that TabPFN not only significantly outperforms Bayesian-optimized Light GBM and Random Forest in prediction accuracy but also improves modeling efficiency by two orders of magnitude (inference time of only 0.62 seconds).Combined with SHAP interpretability analysis,this study quantitatively elucidates the dominant effects of temperature and catalyst,and reveals the flow-rate-enhanced mixing mechanism.This strategy effectively overcomes the data bottleneck in microchemical processes and provides an efficient,general-purpose tool for process modeling.
LiuL, ZhangX, LiX, et al. Study on effect of boric acid on the stability of 1,5-naphthalene diisocyanate-based prepolymer and properties of the elastomers therefrom[J]. Materials Today Chemistry, 2025, 44:102578.
[2]
CarrattG M, ShervinC R, SoongD S. Experimental study and computer simulation of a two-stage continuous polymerization process for the production of methyl methacrylate[J]. Polymer Engineering & Science, 1984, 24(6):442-454.
[3]
ZhangZ, LiT, DongY, et al. Unusual phenomena of ethylene homopolymerization catalyzed by α-diimine nickel emerging in a microflow platform[J]. RSC Advances, 2025, 15(8):5801-5807.
[4]
ChangY, XingY, ShangQ, et al. High time-resolution and data-driven kinetic modeling of cumene autoxidation based on a microreactor system[J]. AIChE Journal, 2025, 71(9):e18873.
[5]
Lopez de ParizaX, ErdmannT, ArrecheaP L, et al. Synthesis of tailored segmented polyurethanes utilizing continuous-flow reactors and real-time process monitoring[J]. Chemistry of Materials, 2021, 33(20):7986-7993.
[6]
SiragusaF, CraneL, StiernetP, et al. Continuous flow synthesis of functional isocyanate-free poly (oxazolidone)s by step-growth polymerization[J]. ACS Macro Letters, 2024, 13(5):644-650.
[7]
PistikopoulosE N, TianY. Advanced modeling and optimization strategies for process synthesis[J]. Annual Review of Chemical and Biomolecular Engineering, 2024, 15(1):81-103.
[8]
MaH, ZhaoQ, YaoC, et al. Effect of fluid viscosities on the liquid-liquid slug flow and pressure drop in a rectangular microreactor[J]. Chemical Engineering Science, 2021, 241:116697.
[9]
JurtzN, KraumeM, WehingerG D. Advances in fixed-bed reactor modeling using particle-resolved computational fluid dynamics (CFD)[J]. Reviews in Chemical Engineering, 2019, 35(2):139-190.
[10]
LiY, CaoH, WangX, et al. A new correlation-similarity conjoint algorithm for developing encoder-decoder based deep learning multi-step prediction model of chemical process[J]. Chemical Engineering Science, 2024, 288:119748.
[11]
XuX, KangQ, ZhangW, et al. An FPM-DM hybrid model for yield prediction of gas-liquid micro-sulfonation[J]. Chemical Engineering Science, 2025, 312:121670.
[12]
ChangY, ShangQ, YanZ, et al. Data-driven models for microfluidics:A short review[J]. Biomicrofluidics, 2024, 18(6):061503.
[13]
ChangY, ShengL, WangJ, et al. A general neural network model co-driven by mechanism and data for the reliable design of gas-liquid T-junction microdevices[J]. Lab on a Chip, 2023, 23(22):4888-4900.
[14]
QinM, LiuY, WangM, et al. Meta-learning driven low-data multi-objective optimization of microchannel gas-liquid sulfonation[J]. Chemical Engineering Science, 2025, 320:122379.
[15]
PengY, WangL. Heterogeneity-aware transfer learning for high-dimensional linear regression models[J]. Computational Statistics & Data Analysis, 2025, 206:108129.
[16]
WangH, PeiJ, YanF, et al. Application of TabPFN model on the energy performance improvement of high-power multistage centrifugal pump[J]. Energy, 2025, 336:138399.
[17]
HollmannN, MüllerS, PuruckerL, et al. Accurate predictions on small data with a tabular foundation model[J]. Nature, 2025, 637(8045):319-326.
[18]
HuangW, ChenX, LvQ, et al. Rapid preparation of waterborne polyurethane dispersions using continuous-flow microreaction technology[J]. Reaction Chemistry & Engineering, 2025, 10:1326-1336.
[19]
ChooplouC A, BalahangS, GhodsianM, et al. Predicting scour depth in the presence of aprons using XGBoost-Optuna[J]. Applied Soft Computing, 2025, 184:113766.
PDF
(1970KB)
