Abstract: The challenges of noise and temperature rise of air-core reactor have always been the hot issues concerned by scholars on account of its long-term exposure to electromagnetic-thermal-force multi-physics fields. In this paper, a simulation model of three-dimensional electromagnetic-structure-sound and fluid-temperature fields is established on COMSOL according to design parameters of reactor, and the results of sound and temperature fields with and without sound-insulation device are obtained. At first, the sound pressure level and temperature rise distribution characteristics are analyzed, and by combining the central composite design and finite simulation results, the results of sound and temperature fields are acquired with different parameters of sound-insulation device. Then, a neural network model is established, which constructs a correlation between the sound pressure level and the hot temperature of the reactor with sound insulation device. Subsequently, the sensitivity analysis technique is accustomed to derive the influence of the structure parameters on the sound pressure level and the hot temperature. Finally, the multi-island genetic algorithm is applied to obtain the optimal parameters of sound insulation device. The optimization results show that the sound pressure level and the hot temperature of the reactor are decreased by 13.9 dB and 26.3 ℃, respectively, under the optimal parameters. Thus, the optimization method has important reference significance for the parameter optimization of sound-insulation device.