EURO-Online login
- New to EURO? Create an account
- I forgot my username and/or my password.
- Help with cookies
(important for IE8 users)
1342. Counterfactual Analysis for Explicability and Fault prevention of Complex Energy Systems
Invited abstract in session TA-27: Counterfactual Analysis Across Diverse Domains, stream Mathematical Optimization for XAI.
Tuesday, 8:30-10:00Room: 047 (building: 208)
Authors (first author is the speaker)
| 1. | Martina Fischetti
|
| 2. | Emilio CARRIZOSA
|
| IMUS - Instituto de Matemáticas de la Universidad de Sevilla | |
| 3. | Juan Miguel Morales
|
| Applied Mathematics, University of Málaga |
Abstract
Energy systems are often very complex systems, where different components interact in a not-trivial manner. In this context, identifying faults in the system and especially identifying their reason can be a complex engineering problem. Take for example an offshore wind farm. As offshore wind turbines can be very expensive to reach, faults are very expensive to handle. Therefore it is very important for energy companies to reliably identify real alarms. For this reason, many companies are nowadays using ML to identify failures. We would like to look a step further: we want to understand how the status of the turbine should be changed not to have a failure. We will use Counterfactual Analysis, once a classifier has been trained, to identify how records should be changed in their features to being classified in the “good” class. Finding counterfactual explanations amounts to solving a mathematical optimization model, where we want to minimize the number of changes (or the total cost associated to changes) to an instance to maximize its probability of being classified as a good instance, subject to different constraints. We will present some preliminary results on the topic
Keywords
- Artificial Intelligence
- OR in Energy
- Machine Learning
Status: accepted
Back to the list of papers