There is a growing interest in applying machine learning methods on large amounts of data to solve complex problems, such as prediction of events and disturbances in the power system. This paper is a comparative study of the predictive performance of state-of-the-art supervised machine learning methods. The event prediction models are trained and validated using high-resolution power quality data from measuring instruments in the Norwegian power grid. The recorded event categories in the study were voltage dips, ground faults, rapid voltage changes and interruptions. Out of the tested machine learning methods, the Random Forest models indicated a better prediction performance, with an accuracy of 0.602. The results also indicated that rapid voltage changes (accuracy = 0.710) and voltage dips (accuracy = 0.601) are easiest to predict among the tested power quality events.

We explore the use of deep reinforcement learning to provide strategies for long term scheduling of hydropower production. We consider a use-case where the aim is to optimise the yearly revenue given week-by-week inflows to the reservoir and electricity prices. The challenge is to decide between immediate water release at the spot price of electricity and storing the water for later power production at an unknown price, given constraints on the system. We successfully train a soft actorcritic algorithm on a simplified scenario with historical data from the Nordic power market. The presented model is not ready to substitute traditional optimisation tools but demonstrates the complementary potential of reinforcement learning in the datarich field of hydropower scheduling.

The power system is changing rapidly, and new tools for predicting unwanted events are needed to keep a high level of security of supply. Large volumes of data from the Norwegian power grid have been collected over several years, and unwanted events as interruptions, earth faults, voltage dips and rapid voltage changes have been logged. This paper demonstrates the application of clustering and dimensionality-reduction techniques for the purpose of predicting unwanted events. Several techniques have been applied to reduce the dimensionality of the datasets and to cluster events based on analytical features, to separate events containing faults from a normal situation. The paper shows that the developed predictive model has some predictive capability when using balanced datasets containing similar muber of fault events and non-fault events. One of the main findings, however, is that this predictive capability is significantly reduced when using unbalanced datasets. Thus, the development of an accurate predictive model based on normal power system conditions, i.e. an unbalanced dataset of events and non-events, is a topic for further research.

No abstract

This paper explores the possibilities of shifting certain household consumer-based loads in time, to reduce unnecessary load peaks to the grid which again can cause challenges for Distribution System Operators (DSOs). Historical measured data on household consumption and the consumption profile of certain household appliances (dishwashers, washing machines and dryers) that can be shiftable in time are being used in an optimization model to investigate the potential that shifting these loads may have on the overall grid consumption of aggregated groups of households. The results indicate that such a model is a viable approach to effectively lower the peak load with respect to these appliances, even with consumer behavior and the inconvenience to perform these shifts accounted for. However, the contribution of the considered household appliances is arguably modest with respect to the total load of the household.