This study applied the artificial neural networks (ANNs) model to the thermal data obtained by suspension ignition and combustion experiment of single peanut shells (PS, millimeter scale) pellet under O-2/CO2 atmosphere. ANN11 was the best ANN model for predicting the relevant parameters of PS combustion. The coincidence between ANN prediction data and experimental data was over 99%. Two modes of biomass pellet ignition were observed: homogeneous ignition of volatiles and hetero-homogeneous ignition of volatiles and char simultaneously. The ignition mode was transformed from homogeneous ignition to hetero-homogeneous ignition when oxygen concentration was 50%. In addition, it was observed that ignition at the bottom emerged first, and then the upper end was ignited, finally generating an envelope flame. This phenomenon occurred when gas flow temperature exceeded 873 K or the oxygen concentration was greater than 50%. The reduction of ignition delay time and internal ignition temperature from 21% to 50% oxygen concentration was more intense than that of 50% to 100% oxygen concentration. Increasing oxygen concentration or temperature resulted in a shorter, brighter, and more stable volatile flame of biomass pellets, which reduced volatile burnout time. Nevertheless, the impact of gas flow rate on biomass combustion was intricate and irregular.