Mapping and understanding humin coking during carbohydrate conversion is crucial for improving solid catalyst systems. However, models for coke mapping are still in need of development. In this study, a lattice Boltzmann method-based back-propagation artificial neural network reduced-order model (ROM) is developed to map the humin distribution during the conversion process. The ROM reveals three configurations of intraparticle coking distribution (surface focus, middle-layer focus, and central focus coking). The experimental feature of surface-focus configuration is the timely decreasing trend in the macroscopic coke accumulation rate, especially under extreme conditions (surface humins/ central humins >10). Catalyst load, pellet size, substrate concentration (LSC), and temperature collectively influence the coking configurations. As the temperature increases (100−160 °C), the configuration with the highest occupancy in the LSC coordinate space changes from central to middle-layer and finally to surface configuration. Increasing the catalyst loading, reducing the particle size, and lowering the substrate concentration under a threshold number (φ LSC ) in LSC space helps prevent the catalyst from working under the extreme surface coking configuration status.