Xiangwei Feng

As renewable penetration in data-center power supply increases, stochastic renewable output can cause tie-line power fluctuations between data centers (DCs) and the utility grid. This paper proposes a privacy-preserving multi-time-scale tie-line power smoothing method for multiple DCs. A two-stage first-order low-pass filter decomposes tie-line fluctuations into high- and low-frequency regulation targets. Server task shifting tracks the high-frequency target, while uninterruptible power supply (UPS) regulation compensates the low-frequency residual under practical energy and power constraints. Second, a federated adaptive proximal policy optimization (Fed-AdaPPO) framework is developed. Proximal policy optimization (PPO) provides stable policy optimization in the continuous action space, and the upper confidence bound (UCB)-guided adaptive exploration improves task-shifting exploration. Critically, only Critic gradients are aggregated across DCs; Actor networks, raw workload data, and user-sensitive information remain local. This design reduces the risk of exposing local state-action mappings. Results show that coordinated server-cluster and UPS regulation reduces the standard deviation of tie-line power by at least 33.4% while maintaining service quality and data privacy.

The application of federated learning in Virtual Power Plants (VPPs) addresses the data silo issue between VPPs and enhances their ability to cope with nonlinear and stochastic scheduling characteristics, which enables VPPs better accommodate distributed energy resources and flexible loads while participating in frequency regulation services. However, although existing federated learning methods strive to solve privacy protection issues, the plaintext transmission of gradients still exposes sensitive data to the threat of curious power control centers and external inference attacks. Therefore, a privacy-protected horizontal federated reinforcement learning approach for multi-VPP optimal scheduling is proposed in this paper. Firstly, a cost-based global optimization scheduling model for multiple VPPs is constructed, modeling the internal scheduling process of VPPs as a Markov decision process. Then, an improved secure horizontal federated multi-VPP collaborative training method is presented, and local models are trained using the Deep Transformer Q-Network algorithm, with local differential privacy and CKKS homomorphic encryption implemented to ensure privacy protection. Finally, a case study is conducted using frequency regulation ancillary service market data and the IEEE-39 bus system structure. Simulation results show that the proposed approach outperforms similar algorithms, achieving high levels of privacy protection and economic operation for VPPs.

In recent years, Electric Vehicles (EVs) have be-come widely used due to their environmental friendliness. How-ever, the increase in essential supporting Electric Vehicle Supply Equipments (EVSEs) has led to a large and diverse cryptograph-ic demand for charging aggregators. The growing demand for computational power cannot be met by local cryptographic com-puting clusters, while cryptographic computing services on the cloud bring risks such as cross-VM (cross-Virtual-Machine) at-tacks. Meeting the requirements of security and timeliness simul-taneously poses challenges for cryptographic service architecture and scheduling algorithms. To address these challenges, a heter-ogeneous task processing architecture is proposed to integrate local and cloud computing resources. Next, a task queue is con-structed to transform the dynamic scheduling problem of multi-ple cryptographic tasks into a queue-jumping sequential deci-sion-making problem. By modeling security requirements as con-straints, the problem is modeled as a Constrained Markov Deci-sion Process (CMDP) formulation. A Deep Reinforcement Learn-ing (DRL) method with recurrent networks and a safe action mechanism are proposed to solve the formulation, which utilizes the temporal characteristics of cryptographic tasks to achieve better scheduling performance under constraints. Finally, com-pared with the state-of-the-art method, our proposed approach is validated through simulation experiments, demonstrating supe-rior security and enhanced processing efficiency across various scenarios.