Federated learning application in distributed energy trading in integrated energy system

Abstract

Privacy protection in electricity market transactions is a long-term topic, and it must be considered in the application of any new power system project today. This paper explores a new welfare optimization method considering privacy protection in the field of energy trading based on federated learning (FL). First, this paper models an integrated energy system (IES) with one distributed network manager (DNM) and multiple load aggregators (LAs). This model adopts the idea of Stackelberg game, makes DNM dominate the trading game, and makes LAs as followers. Then, it uses the FL-Stackelberg method proposed for the first time to solve the established individual welfare optimization model participating in the game and compares the results with those of the traditional hierarchical optimization method and the Mathematical Program with Equilibrium Constraint optimization method. Numerical results at the end of the paper prove the good accuracy and calculation speed of the proposed FL-Stackelberg​ optimization algorithm, and demonstrate the reliability of FL in engineering practice of IES.

Keywords

• Federated learning (FL)
• FL-Stackelberg game
• Integrated energy system (IES)
• Privacy protection