The temperature field distribution of the converter RIP bushing is an important issue that needs to be paid attention to during its operation and maintenance. With the development of digital technology, it is required that the temperature field distribution can be quickly calculated, so that it can be combined with sensor data to achieve the evaluation and diagnosis online. The existing finite element method takes a long calculation time, so this article proposes a data-driven calculation method. Firstly, the temperature field distribution under different boundary conditions is obtained through precise electromagnetic-thermal-fluid coupling calculation by finite element simulation as the basic input dataset. Then, a neural network is used to learn and predict the temperature field distribution in any state. The results indicate that the technical route proposed in this article can control the output error within 1%. The research results can provide reference and ideas for digital online real-time simulation and evaluation of power equipment.