Deep learning-based trustworthy dispatch model for virtual power plants in megacity power grids
Year: 2026
Type: article
Abstract: A deep learning-based trustworthy dispatch model for virtual power plants (VPPs) operating in megacity power grids is proposed to address the interrelated challenges of data scarcity, trust deficits, and operational uncertainty. The proposed framework integrates a hybrid data augmentation module combining a physically constrained generative adversarial network with temporal consistency regularization, a blockchain-anchored multi-dimensional trust evaluation mechanism, and a Wasserstein distance-based distributionally robust optimization (DRO) formulation. A hierarchical dispatch architecture is established, coupling day-ahead DRO with real-time model predictive control, while multi-agent deep reinforcement learning coordinates distributed energy resources under partial observability. Experimental validation is conducted using 18-month operational data from a VPP demonstration project in eastern China, comprising 45 photovoltaic systems, 28 wind turbines, 15 battery storage units, and 120 flexible load aggregators. The proposed model achieves a root mean square tracking error of 0.43 MW, representing a 53% reduction compared with a deep reinforcement learning baseline, and attains a renewable energy utilization rate of 82.4% with only 1.4% physical violation rate in synthetic data. The average daily operating cost is reduced to $9,865, outperforming deterministic, stochastic, robust, and reinforcement learning benchmarks by margins of 30.2%, 21.0%, 33.9%, and 15.7%, respectively. Ablation studies confirm that the DRO formulation contributes the largest performance gain, while the trust mechanism effectively incentivizes reliable participant behavior through transparent economic adjustments. The framework maintains feasibility and cost efficiency under renewable penetration levels up to 70%, demonstrating its scalability and robustness for practical VPP operations in highly urbanized power systems.
Source: Frontiers in Physics
Authors: Yuandong Jiang, Mingyu Ou, Jiangnan Li
Institution: China Southern Power Grid (China)
Language: English
Topic: Smart Grid Energy Management
Subfield: Electrical and Electronic Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $3,077
OpenAlex ID: W7214970554