AI to reduce the cognitive load and support distribution grid operators
The system learns from historical data of outages’ events in the distribution grid, in order to support human operators in the processing of large volumes of information.
25th January 2021
INESC TEC and EDP Distribuição developed a project that resorts to artificial intelligence (AI) to reduce the cognitive load of human operators in the dispatching and control centres of distribution grid, enabling monitoring and real-time analysis of events that take place during grid operation.
The project AI4Substation, “Artificial Intelligence to Reduce the Cognitive Load of Human Operators - Alarm Management in HV/MV Substations”, aims to address one of the major challenges in the operation of high (HV) and medium (MV) voltage distribution grids: the excessive cognitive load that grid control centre operators are subject to, due to the large volume of data (and alarms) generated by all the existing monitoring and protection equipment.
By using techniques from data knowledge extraction and AI, the tools developed throughout the project, by researchers from the Centre for Power and Energy Systems (CPES), seek to identify patterns in the record of events generated in substations, in order to detect potential anomalous behaviours in the protection and automation systems in this type of facilities. This knowledge, extracted from millions of records, is then reorganised and presented in a simple way to human operators, thus facilitating the decision-making process.
The optimization of processes should lead to a better and more proactive operation of the distribution grid, promoting a reduction in the duration and frequency of service interruptions, during different outages’ events in the distribution grid.
From language processing to data analysis
This project led to two computational applications based on AI for analysing contexts in sequences of events generated by the equipment of HV/MV substations.
The first application combines word embeddings (normally used in natural language processing) and clustering techniques to identify similar events (i.e., with similar log messages) in the historical record of high voltage lines. This unsupervised learning approach allows the human operators to classify future occurrences in one of the categories more quickly.
Concerning the second application, developed concurrently, it resorts to the automatic analysis of the historical records generated during alarm occurrences in MV and HV lines, for the autonomous identification of anomalous behaviours, regarding the performance of the protection functions of the line panels.
Collaboration opens the door to pioneering implementation
Although success cases in the application of AI are beginning to emerge, there are no known initiatives to operationalise these concepts in a real context, in order to support human operators’ decisions in grid control centres.
The ongoing collaboration between INESC TEC and EDP Distribuição teams enabled the development of a promising proof of concept. The tools developed within the scope of the project, still in the testing phase, should reach the adequate maturity to be quickly industrialised and integrated into a distribution management system (DMS).