Deep learning enhanced situation awareness for high renewable-penetrated power systems with multiple data corruptions

High renewable penetration and inevitable data corruptions can prominently jeopardise the security of power systems and greatly challenge the conventional situation awareness (SA). This study proposes an enhanced SA model that solves two major difficulties faced by the conventional SA. The first difficulty is to accurately detect anomalies, especially the imperceptible variation of renewable power output. This is addressed by a novel aggregation of random matrix and long short-term memory network. The model's high accuracy and alertness in real-time anomaly detection are achieved by a newly proposed perceptual indicator. The second difficulty is to be robust against multiple data corruptions. In this connection, a dedicated workflow is designed to mitigate the impact of data corruptions from two stages, which ensures the robustness of the enhanced SA model. By comparing with several existing conventional SA models, the proposed enhanced SA model has shown its prominent superiority in several practical scenarios. In addition, a fast security check is also achieved by the enhanced SA model to indicate the security margin of the system on different renewable penetration levels. The enhanced SA model can reinforce the system operators' observability on insecure risks and hedge them against potential data manipulations or cyber attacks.