Machine learning-assisted outage planning for maintenance activities in power systems with renewables

Abstract

The optimal coordination of maintenances is becoming increasingly important to guarantee the security of supply in renewable-dominated power systems. However, current planning tools are plagued with tractability issues arising from the need to comply with operational security standards. The grid must indeed safely accommodate any unexpected contingency occurring during the scheduled maintenances, which requires simulating many different scenarios. To alleviate this computational burden, this paper proposes to leverage machine learning models to predict the outcome of contingency analyses in a fast and reliable manner. The methodology is tested on the full regional transmission grid of Belgium, covering the voltage levels from 150 kV down to 30 kV. Different models, including naive Bayes classifiers, support vector machines and tree-based models, are tested and compared. Outcomes reveal that random forests consistently outperform other benchmarks, by identifying with an accuracy higher than 90% the time periods during which maintenances can be safely performed. Also, we show that the expected rise in renewable generation will impact the maintainability of the future system, with an increase of up to 20% of unsuitable periods to perform maintenances in some grid areas.