<p>As power systems transition toward net-zero emissions, Virtual Power Plants (VPPs) play a crucial role in aggregating distributed energy resources (DERs) to support decarbonized, flexible grid operations. However, a major challenge in VPP implementation is designing real-time dynamic pricing strategies that fairly balance the conflicting economic interests of prosumers, consumers, and energy service providers (SPs). This work addresses this challenge by proposing a model-free Reinforcement Learning (RL) framework that utilizes the Proximal Policy Optimization (PPO) algorithm to dictate optimal pricing policies. We simulate a comprehensive VPP environment encompassing a microgrid of consumers and prosumers (with photovoltaic generation), a shared community battery, dynamic electric vehicle (EV) charging loads, and the traditional power grid. The VPP’s pricing task is formulated as a Markov Decision Process (MDP) aiming to minimize a total cost function. This cost function uniquely relies on a convex combination of normalized costs for all participants, alongside a dynamic penalty for time-varying carbon emissions associated with grid imports. Simulation results demonstrate that the PPO agent effectively learns adaptive pricing strategies that not only balance multi-party profitability and fairness but also actively incentivize local energy balancing. By minimizing nighttime reliance on carbon-intensive grid imports, the proposed AI-driven framework provides a viable pathway for smart local energy markets to achieve economic efficiency and long-term environmental sustainability.</p>