As a crucial component in power systems, the variation in hot spot temperatures of T-type cable terminals significantly impacts the safety and reliability of cable operation. Therefore, accurate prediction of the hot spot temperature in T-type cable terminals is significant for preventing faults. This paper proposes a method for predicting hot spot temperatures of T-type cable terminals that based on temperature field simulation and deep learning. Firstly, a finite element simulation model for the hot spot temperature of T-type cable terminals is established, obtaining training samples of hot spot temperatures under various environmental temperatures and load conditions. Secondly, a fusion prediction network structure based on Physics-Informed Neural Networks (PINNs) and Recurrent Neural Networks (RNNs), referred to as the PINNs-RNN network, is proposed. This deep learning model for temperature prediction of T-type cable terminals achieves high-accuracy prediction of hot spot temperatures. The results demonstrate that the proposed PINNs-RNN deep learning model can achieve prediction of hot spot temperatures in T-type cable terminals with higher accuracy compared to traditional deep learning models, which has significant engineering implications for tracing and preventing thermal faults in T-type cable terminals and ensuring the stable operation of distribution networks.