Fuel cell auxiliary power unit (FC-APU) system for heavy-duty trucks has been attracting much attention as an alternative to engine idling to reduce pollutant emissions. In this system, a reformer, which converts diesel, the main fuel of heavy-duty trucks, into syngas suitable for the fuel cell, is essential. A kinetic model of the reforming can be helpful in tasks such as the reactor design and system operation, but it is highly challenging to develop such a model due to the complexity of the reforming reaction network. Thus, in this study, we propose a systematic approach to build a reduced kinetic model of the diesel autothermal reformer. Specifically, microreactor experiments are conducted using a commercial diesel fuel, where the temperature, oxygen-to-carbon ratio (OCR), and steam-to-carbon ratio (SCR) are varied. Then, based on the composition measurements of the exit gas, optimal kinetic models are constructed by iteratively solving a parameter estimation problem for the candidate kinetic models with varying number of reactions. To reduce the number of candidates significantly, lumping techniques are used to generate reduced kinetic models. A stochastic optimization method is employed for the parameter estimation in which the concept of cross-entropy is used. The optimal reduced kinetic models with 3 and 4 reactions showed a consistency with the literature in terms of the reactions included. The accuracy of optimal reduced kinetic models increased, showing 21% improvement as the number of reactions increased from 3 to 4.