Measurement-Driven Source Tracing of Torsional Subsynchronous Oscillations Caused by Open-Loop Modal Resonance

Abstract

Open-loop modal resonance is the mechanism that a grid-connected wind farm (WF) excites dangerous torsional subsynchronous oscillations (SSOs) of a synchronous generator. However, the open-loop modal resonance analysis (OMA) has to rely on the parametric model, while, in practice, it is often difficult to gain the parametric information of wind farms so as to establish the parametric model for the OMA. Hence, this article proposes a method of the OMA based on the measurement data, rather than parametric model. The proposed method is particularly for the application to trace the source of torsional SSOs, which is the trouble-making wind farm taking part in the open-loop modal resonance. The proposed method of the OMA is developed by the machine learning to train a model of convolutional neural network (CNN). In order to solve the problem of lack of sufficient and labeled training data, it is proposed to build a simplified simulation system. Then, the CNN model is trained by deep transfer learning algorithm using the training data generated by the simulation system to trace the source of torsional SSOs in a practical power system from limited measurement data. In this article, the proposed method is demonstrated and evaluated by a test example.