With the ongoing electrification of society and an increasing number of critical societal functions dependent on the power grid, greater demands are being placed on power quality and grid stability. To meet these demands, both the further development of existing solutions and the creation of new technologies are required. A central question is therefore whether machine learning can serve as a tool to help improve power quality.

This thesis was carried out at University West in Trollhättan and consists of a literature review combined with two practical experiments, with the aim of investigating potential areas of application for machine learning in the context of power quality.

The literature study covered the fundamentals of machine learning, its methods, requirements, and limitations, as well as its use in other industries. The study also examined previous research on the application of machine learning within power quality, along with key problem areas in the field of power quality in general.

The practical component included two experiments: the first aimed to forecast electricity consumption within Power Area 4, based on weather data from four selected cities. Training data covered the years 2022 to 2024, while the testing was conducted using data from 2025. The second experiment focused on identifying voltage sags in synthetically generated voltage samples, using separate datasets for training and testing. These two experiments were chosen to illustrate how machine learning can be used both as a forecasting tool and for classification purposes.

The results showed that the electricity consumption forecasting model could explain 83.79% of the variation in the 2025 test data. The voltage sag identification model achieved an accuracy of 75.00% in determining whether a voltage sag occurred. In addition to the experiments, the study identified a total of thirteen possible areas of application for machine learning within power quality, covering a wide range of relevant challenges. For each area, suggested applications and suitable machine learning methods were presented.

The study is considered to have fulfilled its objective by both identifying relevant application areas and illustrating practical uses of machine learning within power quality. At the same time, it is clear that further work is needed before such solutions can be safely, reliably, and effectively implemented in real-world power systems.