The shell-side fluid heat transfer characteristics of a hairpin heat exchanger have been studied by numerical simulation. In order to improve the comprehensive performance index PEC (the ratio of total heat transfer to total power consumption) and reduce the dimensionless material cost M' (the ratio of heat exchanger material cost to original structure material cost), a neural network model was established. The non-dominated sorting genetic algorithm (NSGA-Ⅱ) was used to optimize the four design variables of the key dimensionless parameters: baffle spacing l', baffle notch height h', curvature radius r' and Reynolds number Re. The results show that within the scope of this study, the heat transfer of the elbow section accounts for 5.0%-16.3% of the total heat transfer of the heat exchanger, while the power consumption only accounts for 0.5%-1.0% of the total power consumption, indicating that the existence of the elbow section structure significantly improves the heat transfer performance of the hairpin heat exchanger with only a slight increase in power consumption. After parameter optimization, the optimal value of l' is 2.50, and the optimal value ranges of h', r' and Re are 0.33-0.45, 0.80-1.30 and 8000-11000, respectively. Two representative solutions were selected from the optimal solution set. Compared with the original structure, the PEC of the optimized structure 1 is increased by 25.12%, and the M' is essentially unchanged. The PEC of optimized structure 2 increased by 17.93%, and the M' decreased by 6.56%, indicating that multi-objective optimization has an obvious advantage in the optimization of the structural parameters of hairpin heat exchangers.