Łukasz Lepak
supervisor: Paweł Wawrzyński
The presentation focuses on using machine learning methods for creating trading strategies using forecasts. The main focus is given to the energy market and the prosumer operating on this market. The prosumer buys energy from the market, sells energy he generated, and also stores energy in batteries for his own needs or for further trading. The goal of the prosumer is to maximize his profit by creating appropriate transaction orders on the energy market - selling for a high price and buying for a low price. The model adopted for this problem is Partially Observable Markov Decision Process (POMDP). The presentation describes this model, defining actions, observations, and rewards for reinforcement learning agent representing the prosumer, and also shows how weather forecasts can be used to help the agent create a better trading strategy. Some constraints regarding the prosumer's operation on the market are also discussed.