The automatic creation of Switching Notes
Before the launch, our internal Artificial Intelligence Center of Excellence (AI CoE) did a few interviews throughout the company to assess the potential of AI-based solutions to solve urgent business challenges. One of the core processes for which there seemed to be a lot of potential for automations is the preparation and execution of switching operations.
Switching in substations is a carefully controlled process, because of obvious safety concerns for people and equipment. In order to maximize a safe working environment, the switching procedure is documented in the form of a series of coordinated actions: the so-called switching note. The preparation of these notes is the responsibility of the dispatchers. This is a time-consuming process, and we see that up to half of the created notes are not executed, due to unexpected changes in the planning.
The Innovation department, the AI CoE and business experts (e.g. dispatchers) decided to develop a solution to automate the creation of the switching notes. After a prior analysis, the following two approaches were selected:
- Rule-based approach
This approach uses well-defined business rules provided by the experts from the control centers to create the notes automatically.
- Machine Learning-based approach
In complex switching cases, where associations between the topology and the necessary actions become less clear, the use of business rules can be challenging. In this case, machine-learning algorithms will be applied to learn those hidden rules and dependencies, and create correct switching notes.
Benefits & Ultimate Vision
The automatic creation of switching notes aims to optimize the time of our dispatchers, by ruling out time-consuming tasks, so they can focus on high added-value work.
The automation of switching notes creation will ensure the uniformity of the notes.
Data Acquisition & Engineering
The current checklist application served as the main data source for analysis. This application has historical data of 15 years with more than 40.000 switching notes at its disposal. The database is supplemented with grid data from the energy management system, and can be improved further by adding real-time SCADA information.
User Friendly Interface
A user-friendly interface is a prerequisite for proper interactions between RCC Dispatchers and the automated solution. This will mainly be of use if the tool fails to create a correct switching note – in this case, the dispatcher will need the ability to correct the note conveniently.
The Proof of Concept that was created throughout the project gives promising results. This pilot project has a database which is linked to the real-time grid topology data, which makes the automatically created switching notes compliant with the current grid topology. When a request is entered, the generation only takes a couple of seconds. This way, dispatchers will save a substantial amount of time.
The pilot is successful in the sense that for some types of switching operations up to 95% of the generated checklists are fully correct. On average, more than 50% of the core of the checklists do not have any mistakes. That shows that the creation of switching notes is potentially automatable.
Unfortunately, the data quality of historical checklists sometimes leads to blind spots for the automation:
- Assets or grid topology can change over time
- Sometimes, historical checklists can be wrong
- Not all cases are covered in historical checklists, obviously only the checklists that have been created in the past exist
In order to reduce the amount of these errors and improve the accuracy of the switching notes creating engine, the pilot was extended to collect feedback from business experts. They have the possibility to give a score to the note or give additional textual feedback. This way, we can improve the quality of the created switching notes by having reliable information of the current correctness and completeness.
Currently the involved stakeholders are investigating how the pilot can be scaled up to a product that can be used in real grid operations based on the results of the project.