Aggregator pricing strategy for community energy management based on multi-agent reinforcement learning considering customer loss or gain

Abstract

Advancements in digitalization have activated power system decentralization, empowering customers to participate in active distribution networks. Forming Local Energy Communities (LECs) with specified regulations can help toward enabling the democratization of future power systems. Many studies have focused on the energy management challenges in this environment based on the role of intermediary players, such as aggregators that are for-profit entities. Nevertheless, the competitive nature of LECs, i.e., customers' capability to select their aggregator, is usually ignored in the economic policy-making problem. This paper proposes a pricing strategy for aggregators, as a techno-economic issue, based on Multi-Agent Reinforcement Learning (MARL) and Long-Short Term Memory (LSTM) forecasting model to enable the required flexibility specified by the utility utilizing the demand response. The upper layer is responsible for the aggregator pricing considering its profit, including the customer loss or gain. On the lower layer, different customers react to the price signals to reduce their bills considering dissatisfaction. The correctness and effectiveness of the model are shown based on the simulation studies of two timeframes on a real residential community dataset, i.e., various smart buildings with flexible loads. The short-term simulations (1 day) show the enabled demand response with different pricing policies. The mid-term simulations (90 days) evaluate the impact of the aggregator’s pricing strategy on its profit and number of active customers. Finally, the aggregator can choose its pricing strategy based on the results of the mid-term evaluation integrating the remaining customers’ value based on the selected pricing range.