Machine Learning Based Trading Strategies in Local Energy Markets

For the economic and efficient energy system design, it is an urge requirement to develop appropriate energy trading models for local markets between the clustered prosumers and distributed energy resources. One of the critical challenges in energy trading is acquiring the information for actions and strategies to make accurate decisions to whom energy to share, sell or trade and on what criteria basis (i.e., number of participants, amount of energy, time horizon, price etc.). Such information-based decision-making actions, and strategies can broadly be classified into mathematical based models and learning models. Previously, the energy trading problem in local markets is considered individually for both mathematical and learning models. As an examples, game theoretical and reinforcement learning approaches are investigated respectively. There are very limited research attempts to employ both approaches simultaneously to improve the decision making on the basis of learning state and to handle large state, transition and action spaces.
In this PhD project, Deep Reinforcement Learning (DRL) is applied to form cooperative coalition to efficiently facilitate energy trading and address related uncertainties for local market especially through minimizing internal energy trading price (IETP), maximizing internal energy trading amount (IETA) and maximizing the financial payout reward (FPR) as objective functions. Moreover, the agent’s interaction in this work is intended to carry out through the iterations of reinforcement learning using the state, action, reward and transition of state and actions (SARSA) algorithm as a property of Markov chain to update actions and state of transition and finally form the optimal coalitions for specific time intervals to improve the overall reward and outcomes of multiple object functions.