The performance of the thermal management system has a great influence on the stability of the special purpose vehicles, and it is of great significance to enhance heat transfer of the radiator. Common research methods for radiators include fluid mechanics numerical simulation and experimental measurements, both of which are time-consuming and expensive. Applying the surrogate model to the analysis of flow and heat transfer in louver fin can effectively reduce the computational cost and obtain more data. A simplified louver fin heat transfer unit is established, and computational fluid dynamics (CFD) simulations is used to obtain the flow and heat transfer characteristics of this geometry structure. A three-factor and six-level orthogonal design is carried out with three structural parameters, the angle θ, the length a and the spacing Lp of the louver fins. The results of the orthogonal design are subjected to range analysis, and the effects of the three parameters θ, a and Lp on the j, f and JF factors are obtained. On this basis, a proxy model of the heat transfer performance for louver fins was established based on the artificial neural network algorithm, and the model was trained with the data obtained by the orthogonal design, and finally the fin structure with the largest JF factor was found. Compared with the original model, the optimized model improves the heat transfer factor j by 2.87%, the friction factor f decreases by 30.4% and the comprehensive factor JF increases by 15.7%.