The hot-spot temperature of transformer windings is a crucial indicator of internal defects. Given that current methods for measuring the hot-spot temperature of transformers are not applicable to those already in operation and suffer from data lag, this paper introduces a novel inversion method combining ultrasonic sensing technology, multiphysics simulation, and the K-Nearest Neighbors algorithm. Leveraging the penetrative ability and temperature sensitivity of ultrasonic sensing, we established a detailed physical field simulation model. This study extensively investigates the characteristics of ultrasonic wave signals inside transformers under different temperature fields, ranging from 40°C to 110°C at 10°C intervals, and various ultrasonic wave emitter conditions. By extracting key features of the acoustic signals, such as peak time, propagation time, and peak amplitude, we successfully achieved accurate inversion of the winding hot-spot temperature. The results demonstrate that this method exhibits a high accuracy rate (98.57%) in inverting the internal winding hot-spot temperature of transformers, offering an efficient and reliable new approach for measuring winding hot-spot temperatures in existing transformers.