A Few-shot Learning Model based on a Triplet Network for the Prediction of Energy Coincident Peak Days
Keywords:Classification, few-shot learning
In an electricity system, a coincident peak (CP) is defined as the highest daily power demand in a year, which plays an important role in keeping the balance between power supply and its demand. Advanced information about the time of coincident peaks would be helpful for both utility companies and their customers. This work addresses the prediction of the five coincident peak days (5CP) in a year. We present a few-shot learning model to classify a day as a 5CP day or a non-5CP day 24-hours ahead. A triplet network is implemented for the 2-way-5-shot classifications on six different historical datasets. The prediction results have an average (across the six datasets) mean recall of 0.933, mean precision of 0.603, and mean F1 score of 0.733.
How to Cite
Copyright (c) 2022 Jinxiang Liu, Laura E. Brown
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.