Detection of Abnormal System- and Operating Behaviour in the Electrical Grid Operation Based on Industrially Proven AI Technology

Abstract:
The detection of anomalies is a requirement in the german IT security act 2.0 which is mandatory for all system operators. The anomaly detection is typically implemented on a communication system level, however the presented approach is focused on the behaviour of the power system in order to further support the existing anomaly detection approaches. An AI based approach using qualitative labelling and autoencoder algorithms to detect anomalies in measurements is presented. Main use cases for this system are anomalies in transformers, infeed of renewable energy sources and systemwide aonomalies. In this paper, the AI approach is described in detail with exemplary results using real-world test data as well as a system architecture and graphical user interface that can be used to deploy the system. The results show that this approach can be used to detect anomalies in measurements that indicate malfunctioning equipment, faulty configurations or malicious manipulations. The operator is informed about the occurrence of an anomaly and is given information to determine the root cause of the finding.