Capturing Demand Response Dynamics in Market Equilibrium Using Mean-Field Theory

Demand response (DR) is one of the most crucial means to address peak-shaving and valley-filling pressures from renewable energy. Considering the increasement of DR participants and the assessment error from the neglection of demand-side price-perception process, the market regulator needs a more efficient and accurate model to examine the market equilibrium. Thus, the paper proposes a Mean-Field Game (MFG)- and Neutral Network (NN)-based bi-level framework. In the upper level, numerous DR participant's interactive gaming can be depicted by the MFG, where they all response to the uniform time-varying prices by load shifting. In the lower level, the real electricity clearing model is used as an engine providing training data, by which a chained deep NN is trained to simulate DR participants price prediction actions. Case studies based on a modified IEEE 118-bus and 200 DR participants are conducted, and the results validate the effectiveness and efficiency of the propose model.