Detecting the Linear and Non-linear Causal Links for Disturbances in the Power Grid

Unscheduled power disturbances cause severe consequences for customers and grid operators. To avoid such events, it is important to identify the causes and localize the sources of the disturbances in the power distribution network. In this work, we focus on a specific power grid in the Arctic region of Northern Norway that experiences an increased frequency of failures of unspecified origin. First, we built a data set by collecting relevant meteorological data and power consumption measurements logged by power-quality meters. Then, we exploited machine-learning techniques to detect disturbances in the power supply and to identify the most significant variables that should be monitored. Specifically, we framed the problem of detecting faults as a supervised classification and used both linear and non-linear classifiers. Linear models achieved the highest classification performances and were able to predict the failures reported with a weighted F1-score of 0.79. The linear models identified the amount of flicker and wind speed of gust as the most significant variables in explaining the power disturbances. Our results could provide valuable information to the distribution system operator for implementing strategies to prevent and mitigate incoming failures.

Automated summary

In this study, we focused on a power grid in the Arctic region of Northern Norway to detect power disturbances and identify significant variables using machine learning techniques. The results showed that linear models achieved the highest performance in fault detection, with flicker amount and gust wind speed being identified as important factors in explaining the disturbances.