Multi-agent deep reinforcement learning for blockchain-based energy trading in decentralized electric vehicle charger-sharing networks

Abstract

With the integration of renewable energy sources into smart grids and electric vehicle (EV) charger-sharing networks is essential for achieving the goal of environmental sustainability. However, the uneven distribution of distributed energy trading among EVs, fixed charging stations (FCSs), and mobile charging stations (MCSs) introduces challenges such as inadequate supply at FCSs and prolonged latencies at MCSs. In this paper, we propose a multi-agent deep reinforcement learning (MADRL)-based auction algorithm for energy trading that effectively balances charger supply with energy demand in distributed EV charging markets, while also reducing total charging latency. Specifically, this involves a MADRL-based hierarchical auction that dynamically adapts to real-time conditions, optimizing the balance of supply and demand. During energy trading, each EV, acting as a learning agent, can refine its bidding strategy to participate in various local energy trading markets, thus enhancing both individual utility and global social welfare. Furthermore, we design a cross-chain scheme to securely record and verify transaction results of energy trading in decentralized EV charger-sharing networks to ensure integrity and transparency. Finally, experimental results show that the proposed algorithm significantly outperforms both the second-price and double auctions in increasing global social welfare and reducing total charging latency.