A proximal policy optimization-based DRL framework for risk-aware peer-to-peer energy trading integrating E-mobility service providers in local energy communities

Local Energy Communities (LECs) enable local supply–demand balancing through peer-to-peer (P2P) energy trading, but the stochastic variability of renewable generation and electric-vehicle (EV) mobility calls for adaptive rather than static coordination. This paper presents a hybrid framework in which a Proximal Policy Optimization (PPO) policy parameterizes a rolling-horizon mixed-integer linear programming (MILP) market-clearing model. Instead of issuing dispatch commands, the policy produces an hour-ahead guidance vector containing two economic signals—a P2P clearing factor and an e-mobility service provider (EMSP) price markup—and flexibility multipliers for stationary batteries, home EVs, and EMSP hubs. These parameters enter the MILP as objective coefficients and bounded limits, ensuring feeder constraints, market rules, EV mobility requirements, and storage physics, while PPO adapts economic trade-offs online. A Dynamic Network Tariff and a Conditional Value-at-Risk term are embedded in both the reward and the clearing objective to internalize network cost and hedge tail risk, preserving the transparency and auditability of the clearing engine. On a 24-hour, 42-node LEC with PV, wind, batteries, home EVs, EMSP hub chargers, and a small hydrogen subsystem, enabling P2P reduces community grid import by up to 18.3% and grid dependence from 91.0% to 74.4%, raises internal P2P exchange to about 1900 kWh, and lowers daily operating cost by 7.6% versus a no-P2P baseline; the PPO-guided clearing attains the lowest grid import and grid dependence at an operating cost comparable to a static MILP. Under 200 stochastic-mobility days, EV trip satisfaction remains at 98.5% with only a 0.24% cost increase, confirming robustness without compromising tractability, auditability, or DSO compatibility.