Abstract: Many data-driven surrogate models have been proposed to alleviate the computational burden of computational fluid dynamics (CFD) simulations. However, building data-driven surrogate models requires an adequate amount of training data. Due to the computational cost, obtaining sufficient training data to construct data-drive surrogate model which aims to solve the CFD computational burden is a dilemma. This paper proposes a data-driven surrogate modeling approach that only uses limited amount of CFD simulation data. The proposed method based on the Interpolation Cold Diffusion Model (ICDM) transforms the prediction of the species mass fraction distributions using boundary conditions into an image-to-image translation task, which translates the source domain species mass fraction distributions to the target domain species mass fraction distributions guided by the target domain boundary conditions. Unlike original denoising diffusion probabilistic model which adds random gaussian noise, ICDM interpolates between distributions in diffusion process which provides more information than the former for reducing the needed amount of the training data. The approach is demonstrated via application to a 2D methane combustion experiment simulated by the CFD. The model trained by 40 data points produces overall normalized mean absolute errors of 1.89% for CH₄ mass fraction, which is even better than the models constructed by the other generation models with 800 data points. Besides, it is 7 times faster than CFD model. This approach requires only a small amount of training data to quickly build an accurate and computationally cheaper CFD surrogate model, providing a basis for deploying virtual sensors, visualizing internal reaction processes, conducting rapid reaction analysis and exploring optimization of reaction processes.