Multi-Frequency Oscillation Source Location Based on STFT and Two-Stage Deep Transfer Learning

Accurately locating multi-frequency oscillation sources is crucial for ensuring the stability of current power systems. This paper proposes a location method for multi-frequency oscillation sources based on Short-Time Fourier Transform (STFT) and two-stage deep transfer learning. STFT is used to obtain the time-frequency distribution of oscillations. The first stage is used to classify whether the oscillation is low-frequency oscillation (LFO) or sub-synchronous oscillation (SSO), while the second stage is used to locate the oscillation sources specifically. Firstly, the active power oscillation signals of all generators in the entire network are subjected to STFT to obtain spectrograms. Then, generate location images by color linear mapping, transforming the oscillation source location problem into an image classification problem. Additionally, transfer learning is employed to transfer image recognition knowledge from other domains to the field of power systems, improving training efficiency and classification accuracy. The proposed method is tested in a 10-machine 39-bus system containing five doubly fed induction generators (DFIG), and the findings reinforce the precision and resilience of the proposed method in the face of noise.