Multi-timescale rolling optimization coordinating heterogeneous EV and building load flexibility in community energy systems with NGBoost-based renewable forecast

The global energy transition towards carbon neutrality has accelerated the development of zero‑carbon communities. While the integration of distributed renewable energy and flexible loads transforms communities into active prosumers, it also introduces significant challenges in managing multi-dimensional uncertainties and economic risks. To address these challenges, this paper proposes a risk-aware multi-timescale rolling optimization framework for community energy systems (CES), that coordinates economic arbitrage with physical reliability. First, a day-ahead stochastic chance-constrained scheduling model is established, incorporating photovoltaic (PV) probabilistic forecast without prior distributional assumptions and capturing the stochastic charging behaviors of electric vehicles (EVs). Second, an intra-day rolling optimization based on a shrinking horizon is developed to mitigate real-time deviations. This stage explicitly models the dynamic and heterogeneous flexibility of EVs and building clusters to exploit arbitrage opportunities in short-term markets. The effectiveness of the NGBoost-based PV probabilistic forecast model and the EV occupancy tracking process is validated through a case study. Results show that the proposed multi-timescale strategy significantly outperforms traditional baseline methods in joint day-ahead and intra-day markets. It can achieve operational cost reductions ranging from 28.99% to 43.61% under varying confidence levels through coordinated optimization of real-time dispatch and heterogeneous flexibility resources.