Institutional Repository
Technical University of Crete
Technical University of Crete
EN
|
EL
| URI: | http://purl.tuc.gr/dl/dias/14390DA3-7818-4440-B911-284EEBA1D7C1 | ||
| Year | 2022 | ||
| Type of Item | Diploma Work | ||
| License |
|
||
| Bibliographic Citation | Antonios Mastorakis, "Development of a competitive autonomous agent for smart grid energy markets", Diploma Work, School of Electrical and Computer Engineering, Technical University of Crete, Chania, Greece, 2022 https://doi.org/10.26233/heallink.tuc.94575 | ||
| Appears in Collections |
Modern electricity markets require real-time sales and purchases, for which many factors must be taken into account to keep pace with market growth rates. In this context, the international Power Trading Agent (PowerTAC) competition simulates a realistic platform for electricity deals and sales, equivalent to real stock exchanges energy such as Nord Pool and EEX. On this platform, various intelligent software agents – brokers (developed by research teams around the globe) compete with each other, with the main purpose of obtaining the maximum possible profit. Each team creates its agent and through various strategies in both retail and wholesale markets is trying to achieve a better combination of buying and selling. Another important part of the competition is that agents aim to obtain a disproportionately high share of the market, resulting in financial losses due to the obligation payment of huge fees to the regulatory authorities. Against this background, this thesis focused on the improvement of an already existing agent, TUC TAC. This particular agent was created by a research team at the Technical University of Crete in 2020, and managed to finish first in PowerTAC that year. Two main changes were carried out to achieve the objective of improving TUC TAC. The first is the addition of a Predictor (forecast factor) for the competition’s wholesale market. By predicting future wholesale market prices, TUC TAC will be able increase its overall profits. We tackled this problem via classical machine learning methods, including Neural Networks. The second major change is the optimization of a Monte Carlo Tree Search algorithm that was already used by TUC TAC for bidding in the wholesale market’s double auctions, via (a) adding the new predictor but also (b) via regulating better the parameters of the algorithm so that selling the same energy amount in the retail market will lead to smaller fines due to the fewer losses in the broker’s energy balance sheet. We conducted extensive simulation experiments to test our modifications and evaluate various versions of our agent in environments of fluctuating difficulty. Our experiments verify the effectiveness of the TUC TAC agent enhancements provided in this thesis.
Copyright © DIAS 2013
