TY  - EJOU
AU  - Li, Duanjiao 
AU  - Li, Jin 
AU  - Chen, Xiaokai 
AU  - Wu, Hongyao 
AU  - Xie, Haonan 
AU  - Jiang, Bo 
AU  - Liu, Yanmin 
AU  - Qiu, Jian 

TI  - Digital Twin-Based Collaborative Energy Scheduling of Source-Grid-Load-Storage in Multi-Scenario Parks
T2  - Energy Engineering

PY  - 
VL  - 
IS  - 
SN  - 1546-0118

AB  - Renewable intermittency and load uncertainty cause significant operational challenges for park-level source-grid-load-storage (SGLS) systems. Existing digital twin scheduling studies mainly focus on static modeling, lacking closed-loop integration between high-fidelity uncertainty scenario generation and real-time decision-making, while conventional uncertainty-aware approaches suffer from inaccurate virtual-physical mapping and time lags in multi-timescale optimization. To address these issues, this paper proposes a digital twin-based collaborative energy scheduling method with three key innovations. First, a five-dimensional digital twin architecture is constructed to achieve hierarchical mapping and bidirectional real-time interaction, reducing virtual-real mapping deviations. Second, a denoising diffusion probabilistic model enhanced with a self-attention mechanism is developed for high-fidelity scenario generation, capturing long-term temporal dynamics of source-load uncertainty. Third, a model predictive control (MPC)-based rolling optimization model is established, using the generated scenarios as prediction inputs with feedback correction to eliminate multi-timescale time lags. Experimental results on a digital twin platform under 60 operating scenarios demonstrate that the proposed method reduces the autocorrelation function (ACF) error to 0.30, lowers total operating cost by 10.9% compared with the strongest baseline, and maintains voltage deviation below 0.04 p.u. under high-volatility conditions, significantly outperforming state-of-the-art benchmarks.
KW  - Digital twin; source-grid-load-storage; self-attention mechanism; denoising diffusion probabilistic model; scenario generation; model predictive control rolling optimization

DO  - 10.32604/ee.2026.083589