In this paper, a novel energy pricing and sharing strategy is proposed to solve the energy management and market bidding problems of a virtual energy station (VES) in multi-carrier energy systems. First, the energy cell-tissue-based group interaction framework and the multiagent-based communication and control architecture are established in the VES. Second, the general model of an energy cell is built, in which the integrated demand response process and the automatic generation control mode of a combined heat and power unit are designed. Third, according to the classification of energy cells, the aggregation rules of the VES without the participation of third-party agents are proposed, and the internal trading mechanism of the VES led by a higher energy cell is designed based on a Stackelberg game. Moreover, a three-stage hybrid stochastic robust optimization strategy is proposed to address uncertainties in market prices and renewable energy output. Finally, the optimization model is linearized and solved by duality theory, KKT conditions, and column-and-constraint generation (CC&G) algorithm. Simulation results prove the rationality and effectiveness of the framework and business model for the VES.

The growing adoption of electric vehicles (EVs) has placed significant demands on power grids, necessitating coordination between EV charging and power dispatching. This paper proposes a novel EV charging system using Multi-Agent Large Language Models (LLMs) to enhance recommendations, optimize decision-making, and dynamically adapt to user behaviors and grid conditions. The system includes a User Agent and an EV Charging Station (EVCS) Agent, connected through a Negotiation Platform for secure data sharing. The User Agent provides personalized recommendations based on historical data, while the EVCS Agent adjusts prices in real time using fine-tuned LLMs. A Conditional Generative Adversarial Network (CGAN) model is used to generate user behavior and pricing data to fine-tune the LLMs. The proposed system effectively adapts to dynamic user behaviors and grid conditions by combining Multi-Agent coordination with fine-tuned LLMs and CGAN-generated data. A case study demonstrates the system’s ability to balance user preferences with power dispatching, offering scalable, efficient, and intelligent solutions for modern EV ecosystems.

Thanks to the uncertainty and fluctuation of wind power, the mismatch between the supply and load is increasing. Consequently, wind power should be responsible for the imbalance caused by its volatility. Meanwhile, thermostatic load (TCL) is becoming a promising regulation capacity provider with the development of demand side technology. This paper proposes a combined energy and regulation market framework to encourage wind power to pay for the regulation market cost. And a cooperation framework is presented, in which wind power and two types of TCLs exchange energy and regulation capacity to obtain cooperation benefits. The asymmetric Nash bargaining (ANB) is applied to allocate cooperation profits among members. To ensure the autonomy of each member, a decentralized method based on ADMM is introduced. Case studies show that the proposed bidding and trading strategy can increase the profit of each member and lock more regulation capacity in day‐ahead stage, and the proposed allocation method can distribute cooperation profits fairly.

Coupling effects in road-power coupling networks (RPCNs) attract increasing attention, as they are crucial for addressing the road-power intertwined challenges caused by the rapid growth of electric vehicles (EVs). To solve these challenges, this paper, based on Graphormer and Joint-Embedding Predictive Architecture (JEPA), proposes a road power foundation model (RPFM) as a generic solution. Our RPFM, by employing graph-pretraining methods to bridge RPCNs, demonstrates exceptional performance in downstream tasks such as EV Charging Station (EVCS) Load Prediction, Road Traffic Prediction, and EVCS Location Planning. By experimenting on real world dataset, the proposed methodology is proved to be generic and achieves state-of-the-art performance across downstream tasks by improving an average of 7.53% of baseline models’ performance.