Accurate monitoring of the power equipment or personnel positions is essential for achieving safe operations and effective maintenance in the power system context. Traditional positioning methods mainly rely on Global Navigation Satellite Systems (GNSS) for mea­surement. Whereas under complex and dynamic channel environments affected by electromagnetic interference and frequency-selective fading caused by terminal mobility in indoor and outdoor settings, the per­formance of the GNSS measurement method degrades. To tackle this problem, this paper proposes a multi-mode fusion positioning method that utilizes residual and line-of-sight (LOS) signals in the local refer­ence frame and the 5G network positioning results to input into the Transformer-based Deep Neural Network (T-DNN) which outputs position correction. The results demonstrated that the proposed mechanism possesses the capability of representation learning, which can capture the key features and complex relationships in multi-mode fusion obser­vations from the power system scenarios, and demonstrate satisfactory accuracy in position monitoring and correction.