A probabilistic machine learning approach to detect severe sub synchronous oscillations in DFIG based wind farms

The dynamic properties of the power grid have become increasingly complex due to the continuous transition of the power grid towards the use of renewable energy sources for power generation. The emerging sub synchronous oscillation (SSO) conditions owing to the high penetration of renewable energy resources is one such issue and should be investigated rigorously to maintain the reliability and safety of the grid. Although wide area measurement system can detect instabilities in general, it is inadequate to detect the sub synchronous oscillations. Recent events that have occurred in USA and China bear evidence to this inadequacy. This article proposes a technique for the real time monitoring of the sub synchronous oscillations in DFIG based wind farms and early warning of severe SSO using an artificial intelligence-based approach. Accordingly, a probabilistic Naive Bayes algorithm is trained to anticipate severe SSO that could potentially disconnect wind turbines from the grid, thereby allowing the power system operators to take necessary actions to control the SSO without letting the disconnection of large number of wind turbines from the power grid. The proposed Naive Bayes classifier has a F1-Score, training time and prediction time of 0.97, 0.891 seconds, and 0.0238 seconds respectively. Furthermore, it can provide the confidence level of its prediction rather than a deterministic output as proposed in existing literature. The simulation results demonstrate the effectiveness of the proposed approach for SSO identification.