Search

Browse

• Communities
• Authors
• Year
• Titles
• Subjects

My Space

• Login

Energy consumption prediction via machine learning algorithms

Kontos Stefanos

Full record

---

Summary

As time goes by renewable energy usage in the residential market rises. Electric vehicles usage is in a remarkable upside and this constitutes enough excuse for researchers to invest in upgrading the electricity grid. For that purpose, the Power TAC competition provides a multi-agent simulation platform for electricity markets. In this platform, adversary brokers compete into buying and selling energy. Their primary target is to obtain max profit. The platform consists of retail, wholesale, balancing and tariff markets, which push the complexity of the broker's strategy up. Teams need to construct a broker that can manoeuvre with flexibility among consumers, producers and markets in order to accumulate profits. One such agent was TUC-TAC 2020, the agent that represented the Technical University of Crete in the PowerTAC 2020 international competition, and which was developed by a team of students in which this thesis author participated. TUC-TAC was crowned the PowerTAC 2020 champion, competing against 7 other agents representing universities from 6 different countries. In this thesis, we present TUC-TAC's energy consumption predictor module. The goal in our thesis was to predict the consumption of our agent's customers for the future timeslots. The problem was mainly approached with Machine Learning as a regression problem. Neural Networks were also implemented and tested. The predictor module is integrated to the agent in order to provide information useful for the decision making in the tournament environment. All the different approaches are presented in detail with experimental results and comparisons. We believe that this work can serve as a departure point to build even more successful trading agents in the future.

Available Files

Services