Wholesale Bidding Approaches of an Autonomous Trading Agent in Electricity Markets

Abstract : Energy transition policies, including Energiewende in Germany, plan to replace conventional power plants with renewable energy resources thanks to advances on solar and wind technologies. However, shifting from conventional production to highly volatile renewable energy production will bring a number of challenges as well. One of those challenges is price forecasting accuracy in renewable energy driven markets. Intermediary power actors, e.g., retailers, suffer from that problem at most, due to their load-balancing role (financially). This work presents a number of electricity price forecasting approaches, and aims to refine the forecasting error, using additional drivers such as weather observation data. The proposed bidding approaches have been tested in our broker agent which competes in Power Trading Agent Competition (Power TAC). In detail, our model firstly makes price predictions out of historical market clearing prices, using a hybrid model. This model alters a seasonal regression model by replacing the aged terms with a belief function. Then, those predicted prices are reassessed by means of correlating the weather observations and market clearing prices. The price-driven methods were tested in the Power TAC simulation framework whereas the others use real-world data. Results show that weather data can successfully reduce the forecasting error up to a certain degree. After all, the paper aims to create generic data-driven forecasting models which can be deployed in autonomous trading agents.