ML-Assisted Sub-synchronous Oscillation Detection and Localization in Type-4 Wind Farms under Weak Grid Conditions

Recently, incidents of the sub-synchronous oscillation (SSO) occurrence have significantly increased in the power system integrated with renewable energy resources. In the current power system, as more distributed renewable energy resources replace conventional synchronous generator-based sources, the grid is transformed into a low-inertia power system. The low inertia and distributed nature of renewable energy systems contribute to weak grid conditions. In the type-4 wind turbine generators (WTG) SSO can originate due to the interaction of fast dynamics of power converter’s control and weak AC grid. If SSO is not detected and mitigated in a timely manner, it can cause severe damage to the WTG’s shaft, turbine structure, and can pose severe type of instability in the power system that may lead to cascaded tripping. Therefore, this paper presents a machine learning (ML) based scheme for fast and accurate detection of SSO and localizing the WTG’s control parameter that triggers SSO under the influence of weak AC grid. The proposed SSO detection and localization scheme features a dual neural network structure (DNNS) based on long short-term memory (LSTM). The first NN structure detects SSO and triggers second NN structure when SSO is detected. The second multiclass NN identifies the control parameter of WTG that triggered this SSO and provides a recommendation/ reference for the SSO mitigation scheme. The effectiveness of the proposed ML based SSO detection and localization scheme is verified via confusion matrix and simulation that analyzes different cases related to the ML scheme validation.