The hot-spot temperature of oil-immersed transformers is crucial for their reliable operation. Traditional monitoring methods have limitations like invasiveness or inaccurate calculations. This paper presents a new hotspot temperature inversion algorithm using Physics-Informed Neural Networks (PINN). It combines physical laws with deep learning to improve temperature prediction accuracy. The model takes coil current, top oil temperature, and load rate as inputs and predicts the hot-spot temperature. A custom loss function, integrating physical constraints and traditional data loss, ensures predictions match the transformer’s physics. Simulation experiments under different operating conditions validated the model. Its performance was evaluated using Mean Squared Error and Coefficient of Determination. The PINN-based model outperformed traditional algorithms such as SVR and Random Forest. Results show the PINN model offers a more reliable real-time hot-spot temperature monitoring solution. Prediction errors are consistently within 3 K (Kelvin), making it a promising tool for ensuring the safe operation of oil-immersed transformers in power systems.