Prediction and early identification of evolving faults in electrical systems using machine learning driven methods

Important results from the project

During the project, a prototype based on machine learning (ML) for predicting incipient faults in electrical systems was successfully completed. In addition to the method for predicting incipient faults, Eneryield developed methods for detecting and classifying anomalies, which is value-creating in itself but also fundamental for accurate prediction. The project has shown that ML-based solutions can contribute to increased reliability. Follow-up is planned to be carried out during 2022.

Expected long term effects

The project has highlighted the benefits of ML, as well as the existing conditions and challenges, when applied to power grids The work has deepened the understanding of how different anomalies/disturbances in electricity grids are linked, and what sequences that implies an incipient more serious fault. A plan for implementation and market introduction is set out, and the expected effects on society will in the long run be reduced interruptions and faster troubleshooting, where a minimum of kWh is lost.

Approach and implementation

The work went well. However, data collection was time consuming, but luckily enough time was dedicated from the beginning. Furthermore, to optimize the usefulness, the focus was shifted away from new hardware. The planning of a follow-up project began during the middle of the project and has taken place in parallel with the rest of the work. Focusing the ongoing work on the follow-up at a relatively early stage, benefited both the current project and the long-term development of the innovation.