Hydrogen energy has been considered as one of the solutions to achieve the net-zero emission scenario by 2050. Steam methane reforming is a widely used industrial process for producing hydrogen from natural gas or methane nowadays. Considering that methane could be utilized as a suitable carrier for hydrogen energy, it is anticipated that steam methane reforming will still play an important role in the future energy sector when it comes to hydrogen production, storage, and transportation. In this work, a one-imensional dynamic model is established to simulate the performance of an auto-thermal reforming reactor, which allows for capturing the localized phenomena inside the reactor over time. A set of input parameters is selected based on the Latin Hypercube Sampling method to generate the training data for the surrogate model development. Singular value decomposition and Gaussian Process regression are then implemented on the training data to construct a surrogate model of the reformer. This surrogate model is subsequently utilized in the optimization process to enhance hydrogen production and lower the maximum catalyst temperature within the reactor. The results show that the surrogate model, developed by using singular value decomposition and Gaussian Process, exhibits a high level of accuracy when compared to the physics-based reformer model. Furthermore, the optimization framework built upon surrogate modelling offers the potential to substantially reduce the computational expenses associated with the optimization process, while preserving the precision of the optimization results. This method could efficiently serve as a tool for parameters optimization of such reactors and could be used to guide the operation of these systems toward improved performance.