The days when "voltage" in the power grid had a purely physical meaning are over. At least since Marc Elsberg's thriller Blackout, we know how "tense" the work of grid operators is today.
The danger of fiction becoming reality seems greater than ever. It does not even take a Cyber attack to increase the tension in the control rooms. Connecting more and more volatile infeeds from distributed energy resources challenges the power grids. In addition, electric vehicles and heat pumps on the consumer side are reducing the constancy and predictability of days gone by. Managing volatile load and operating scenarios requires real-time monitoring of the equipment. This enables quick responses and to counteract proactively slowly developing problems.
Systematic Network Monitoring
The PSIdetect software product developed jointly by E.ON and PSI uses Artificial Intelligence (AI) to detect anomalies in network operation at an early stage. This makes systematic grid monitoring possible for the first time. It's tasks are detecting abnormal behavior of feeders and transformers, both separately and in a system context.
Let us take a closer look: The control system provides the measurements.
“In order to train PSIdetect, training data such as historical and synthetically generated data as well as current process and weather data during operation are required,” This data is used to continuously determine the target state and compare it with the actual state."
In order to train PSIdetect, training data such as historical and synthetically generated data as well as current process and weather data during operation are required.
Traffic Light Dashboard Provides a Quick Overview
The software automatically calculates an “anomaly score” that indicates the condition of a piece of equipment or the system. For example, is something vibrating or heating up where it should not be vibrating or heating up?
Based on a permanent comparison with historical data and defined threshold values, PSIdetect identifies and displays even minimal changes and deviations. The traffic light display makes it possible to assess the situation at first glance by showing three statuses:
- 1 (= green) symbolizes perfectly functioning operation.
- 0 (= yellow) means a weak anomaly or potentially imminent technical problems.
- -1 (= red) stands for a strong anomaly or an existing fault.
The system manager in the network control room automatically receives messages about the status.
The analysis starts after detecting and alarming abnormal behavior to the operators:
- Is the deviating sensor value measured correct or is the sensor defective and has triggered a false alarm?
- Was a value incorrectly parameterized during a system changes?
- Has a piece of equipment been tampered with?
- Or is it simply an exotic grid state due to an unusual external situation, such as a solar eclipse.
A comparison with the operating log of the control system is also helpful.
Probability of Occurrence and Pattern Recognition
What does AI do to detect abnormal situations? "We take two aspects into account in our AI," says Dalhues.
"Firstly, the software carries out a physical evaluation of the current equipment status, for example of a transformer. A probability distribution for the high-voltage side of the transformer is stored for this purpose. In other words: What measured values can typically be expected? This makes it possible to determine how likely an operating state is. If an outlier is measured, its probability of occurrence can be evaluated. This can be done in parallel for several dimensions such as voltage, active power and reactive power. The parameters can be set in relation to each other and thus a comprehensive evaluation can be carried out.
On the other hand, the software is capable of recognizing patterns, i.e., comparing temporal progressions and thus pointing out creeping changes."
Each anomaly search gets a time stamp to make processes transparent. The detected anomaly is geographically located on a network map so that network operators are immediately aware of where maintenance staff needs to go to correct a technical problem. Once the system has been activated, the control room personnel can view various details, such as a list of all the equipment being monitored for anomaly detection. Current anomaly scores are also displayed, as well as statistical data to better categorize conditions.
Simple Adaptation through Individual Parameterization
The software can be parameterized according to individual sensitivity properties. This means that notifications can be generated at lower threshold values, not just at -1. The system as a whole can be switched on and off, for example, in the event of a major fault in the grid, to prevent unnecessary error messages from being generated. The AI is usually trained cyclically, but can also be triggered manually if, for example, a large PV system is new to the grid and this is expected to have a major impact on grid management.
Anomaly detection offers a specific user interface and is fully integrated into the Java-based PSI platform. Administrators in the back office, service staff and system administrators receive network status and anomaly information and can make parameterizations.
Building block for compliance with the IT Security Act
"PSIdetect reliably detects relevant deviations in the power grid, i.e., anomalies that can be attributed to Cyber attacks, technical faults or other influences. This enables protective intervention at the earliest time and thus increases the safety and reliability of the power supply," summarizes Stefan Dalhues.
The software is therefore also an important building block for compliance with the IT Security Act, which will require grid operators to use anomaly detection for the control system level from May 2023. The target group is all grid operators with suitable measurement data. These can be municipal utilities, distribution grid operators or transmissiwon grid operators. PSIdetect can also be used in area or industrial networks.