Building Integrated Photovoltaics (BIPV) with Energy Storage Systems (ESS) enables buildings to play a crucial role in on-site PV consumption. However, due to the uncontrollability of PV, buildings often struggle to fully utilize it in real time. This paper proposes a decentralized cooperative power dispatch approach based on multi-agent proximal policy optimization (MAPPO) for cluster consisting of multiple BIPV with ESS. To acquire reliable strategies, a digital twin (DT) is employed as a sample and training environment for MAPPO to minimize cumulative grid power replenishment. An example of a small-scale building cluster is used to demonstrate the coupling of MAPPO and DT. The decentralized dispatch strategy is obtained with a one-hour time step. Verification results indicate a 9.85 MWh boost in PV self-absorption compared to a self-generating self-using strategy. Leveraging DT opens up further possibilities for applying MAPPO to power dispatch challenges.
Keywords Multi-agent Proximal Policy Optimization, renewable energy, digital twin, decentralized dispatch, building integrated photovoltaics