CVaR-Constrained Deep Reinforcement Learning for Low-Carbon Wind-PV Dispatch in ADNs

With the penetration of renewable energy in the power grid such as wind and photovoltaic (PV) energies, their intermittency and variability have significantly posed challenges for the stability and economic benefits of the power grid. It is crucial to optimize the model of power dispatching of wind and photovoltaic energies for ensuring the stability and economic emission of the power grid. To address the problem, we propose an optimization model of wind-PV dispatching that combines deep reinforcement learning and CVaR risk analysis (CVaR-DRL). The proposed CVaR-DRL model integrates CVaR constraints to optimize the utilization of renewable resources, storage dispatch, and thermal generation scheduling. Simulations on the IEEE-33 node system demonstrate that our CVaR-DRL approach is superior to these baseline methods in terms of energy utilization efficiency, peak-hour cost reduction and policy stability. Moreover, it effectively leverages wind energy during low-PV winter conditions, ensuring economic-emission wind-PV dispatching. The results underscore the potential of CVaR-DRL in enhancing renewable energy integration and economic carbon emissions.