Abstract: The temperature rise of transformer is the key factor affecting its operating state and service life. In order to obtain the temperature distribution of an oil-immersed transformer in the winding area accurately, the electromagnetic-fluid-thermal coupling process of the winding is numerically studied by the finite element method (FEM). According to the distribution characteristics of electromagnetic and fluid-temperature field, the calculation models of transformer and winding are established, respectively. Based on heterogeneous grid node data mapping method, the non-average winding loss obtained from the electromagnetic field analysis is loaded into the fluid-temperature field grid as the heat source, and the heat source is modified according to the temperature at each iteration. The bidirectional coupling between electromagnetic field and fluid-temperature field is realized. On this basis, the response surface of the hot pot temperature and winding structure parameters is established, and the multi-objective optimization research with the purpose of minimizing the winding hot spot temperature and conductor material consumption is carried out. The Pareto optimal solution is obtained by the non-dominated sorting genetic algorithm (NSGA-Ⅱ), and the four optimization schemes on the Pareto front are analyzed. The results show that the optimized winding hot spot temperature and the amount of conductor are significantly reduced. The research of this paper provides a feasible method for the optimal design of transformer structure.