We present the design, implementation, and analysis of autonomous brokers for a smart grid ecosystem, in the context of PowerTAC. PowerTAC simulates the real-world smart grid ecosystem through wholesale, retail, and balancing markets. The complexity of grid operations makes designing broker components using artificial intelligence (AI) inevitable. We begin with the challenges prevalent in devising such autonomous brokers. Then, we describe the design, implementation, and performance of our AI-based autonomous agent VidyutVanika (VV). We discuss two variants of VV: VV18 and VV21, which were successful in the annual PowerTAC tournaments, 2018 and 2021, respectively.
In the PowerTAC wholesale market, the brokers purchase electricity for future slots through periodic double auctions (PDA). We first characterize theoretical Nash equilibrium (NE) bidding strategies under certain technical assumptions. In practice, VV18 formulates the PDA of the wholesale market as a Markov decision process (MDP) and uses dynamic programming (DP) to solve it, which is shown to approximate the theoretical NE as well as outperforming several other wholesale strategies presented for the PowerTAC. However, VV18 and other previous wholesale strategies did not consider the supply curve of the power generating company (GenCo) in their bidding strategies. VV21 follows the supply curve of the supplier for every delivery timeslot and uses it to determine a risk-adjusted price band to place bids in the wholesale market, which is shown to reduce the overall unit cost of procurement.
PowerTAC simulates thousands of prosumers in the retail market who subscribe to one of the available tariffs published by brokers. VV18 models the tariff market operation through an MDP and uses Q-Learning to solve the MDP. The VV18 tariff module formulation did not have an explicit way to control capacity transaction penalties, which is remedied in VV21, whose tariff module is designed to maintain a market share within a specified range. The tariffs offered follow a specific price pattern introducing surcharges to reduce energy usage during peak hours which helps in mitigating demand peaks and thus achieving a healthy cash position. We conclude the chapter by enumerating the further challenges in designing different components of AI-based smart brokers.
