hr Hybrid classification-regression metric for the prediction of constraint violations in distribution networks

The massive integration of renewable energy sources in distribution grids is bringing out significant new challenges. Identifying and preventing grid constraints violations is of upmost importance to guarantee the stability and efficient operation of the system. However, grid constraint violations prediction is still a rather unexplored subject, which is still relying mostly on power flow calculations based on often inaccurate consumption and generation predictions. Besides the need for new models that are able to reach effective constraint violation predictions, there is also a need for new result evaluation metrics that are suitable for a problem with particular characteristics. On one hand it is necessary to identify the moments in time in which constraint violations are expected (classification problem) and on the other hand, to predict the constraint violation amplitude (regression problem). This work proposes a new hybrid classification-regression metric that enables capturing the results of both problems simultaneously. The proposed metric identifies the observations that are correctly or incorrectly classified as constraint violations, while converting the classification results into a continuous value that is affected by the quality of constraint amplitude prediction. Results using a distribution network with real consumption and generation profiles show that the proposed metric enables a more effective representation of the results when compared to traditional classification and regression evaluation metrics.

Automated summary

The integration of renewable energy sources in distribution grids poses new challenges and the prediction of grid constraint violations is still an unexplored subject. This work proposes a new hybrid classification-regression metric that effectively represents the results compared to traditional metrics.