Although multi-agent reinforcement learning (MARL) has shown its success across diverse domains, extending its application to large-scale real-world systems still faces significant challenges. Primarily, the high complexity of real-world environments exacerbates the credit assignment problem, substantially reducing training efficiency. Moreover, the variability of agent populations in large-scale scenarios necessitates scalable decision-making mechanisms. To address these challenges, we propose a novel framework: Sequential rollout with Sequential value estimation (SrSv). This framework aims to capture agent interdependence and provide a scalable solution for cooperative MARL. Specifically, SrSv leverages the autoregressive property of the Transformer model to handle varying populations through sequential action rollout. Furthermore, to capture the interdependence of policy distributions and value functions among multiple agents, we introduce an innovative sequential value estimation methodology and integrates the value approximation into an attention-based sequential model. We evaluate SrSv on three benchmarks: Multi-Agent MuJoCo, StarCraft Multi-Agent Challenge, and DubinsCars. Experimental results demonstrate that SrSv significantly outperforms baseline methods in terms of training efficiency without compromising convergence performance. Moreover, when implemented in a large-scale DubinsCar system with 1,024 agents, our framework surpasses existing benchmarks, highlighting the excellent scalability of SrSv.

Recent advancements in Large Language Models (LLMs) and Reinforcement Learning (RL) have shown significant promise in decision-making tasks. Nevertheless, for large-scale industrial decision problems, both approaches face distinct challenges: LLMs lack real-time long-sequence decision-making capabilities, while RL struggles with sample efficiency in vast action spaces. To bridge this gap, we propose Agents Co-Evolution (ACE), a synergistic framework between LLMs and RL agent for large-scale decision-making scenarios. ACE introduces a dual-role trajectory refinement mechanism where LLMs act as both Policy Actor and Value Critic during RL’s training: the Actor refines suboptimal actions via multi-step reasoning and environment validation, while the Critic performs temporal credit assignment through trajectory-level reward shaping. Concurrently, RL agent enhance LLMs’ task-specific decision-making via prioritized experience replay. Through extensive experiments across multiple power grid operation challenges with action spaces exceeding 60K discrete actions, ACE demonstrates superior performance over existing RL methods and LLM-based methods.

Virtual Machine Rescheduling (VMR) plays a crucial role in maintaining service quality and resource efficiency in elastic cloud computing. However, existing datasets and benchmarks primarily focus on VM scheduling tasks, while lacking industrial-scale datasets and standardized evaluation for the more complex and crucial rescheduling problems. To address these challenges, we present IVMR suite, the first industrial-scale suite for VMR research comprising two core components: 1) IVMR-D, an industrial-grade VMR dataset mined from a real cloud data center, integrating complete resource specifications and complex operation constraints. The dataset is systematically structured based on data size and optimization objectives. 2) IVMR-B, a benchmark for the VMR problem that establishes seamless integration of consistent evaluation and the provision of baselines spanning optimization, metaheuristic, heuristic, and machine learning-based methodologies. Our comprehensive experimental evaluation demonstrates that all tested VMR algorithms struggle to effectively balance solution quality with computational efficiency while showing limited scalability across tasks with varying complexity levels. These findings emphasize the urgency of improving VMR algorithms for industrial deployments.

Safe Reinforcement Learning (RL) is crucial for achieving high performance while ensuring safety in real-world applications. However, the complex interplay of multiple uncertainty sources in real environments poses significant challenges for interpretable risk assessment and robust decision-making. To address these challenges, we propose Fuz-RL, a fuzzy measure-guided robust framework for safe RL. Specifically, our framework develops a novel fuzzy Bellman operator for estimating robust value functions using Choquet integrals. Theoretically, we prove that solving the Fuz-RL problem (in Constrained Markov Decision Process (CMDP) form) is equivalent to solving distributionally robust safe RL problems (in robust CMDP form), effectively reformulating the min-max optimization problem into a tractable CMDP with Choquet-integrated value functions. Empirical analyses on safe-control-gym and safety-gymnasium scenarios demonstrate that Fuz-RL effectively integrates with existing safe RL baselines in a model-free manner, significantly improving both safety and control performance under various types of uncertainties in observation, action, and dynamics.

With the increasing penetration of inverter-based renewable energy resources, deep reinforcement learning (DRL) has been proposed as one of the most promising solutions to realize real-time and autonomous control for future carbon-neutral power systems. In particular, DRL-based frequency control approaches have been extensively investigated to overcome the limitations of model-based approaches, such as the computational cost and scalability for large-scale systems. Nevertheless, the real-world implementation of DRLbased frequency control methods is facing the following fundamental challenges: 1) safety guarantee during the learning and decision-making processes; 2) adaptability against the dynamic system operating conditions. To this end, this is the first work that proposes an Adaptive and Safe-Certified DRL (AdapSafe) algorithm for frequency control to simultaneously address the aforementioned challenges. In particular, a novel self-tuning control barrier function is designed to actively compensate the unsafe frequency control strategies under variational safety constraints and thus achieve guaranteed safety. Furthermore, the concept of meta-reinforcement learning is integrated to significantly enhance its adaptiveness in non-stationary power system environments without sacrificing the safety cost. Experiments are conducted based on GB 2030 power system, and the results demonstrate that the proposed AdapSafe exhibits superior performance in terms of its guaranteed safety in both training and test phases, as well as its considerable adaptability against the dynamics changes of system parameters.

Decarbonization of global power systems significantly increases the operational uncertainty and modeling complexity that drive the necessity of widely exploiting cutting-edge Deep Reinforcement Learning (DRL) technologies to realize adaptive and real-time emergency control, which is the last resort for system stability and resiliency. The vulnerability of the DRL-based emergency control scheme may lead to severe real-world security issues if it can not be fully explored before implementing it practically. To this end, this is the first work that comprehensively investigates adversarial attacks and defense mechanisms for DRL-based power system emergency control. In particular, recovery-targeted (RT) adversarial attacks are designed for gradient-based approaches, aiming to dramatically degrade the effectiveness of the conducted emergency control actions to prevent the system from restoring to a stable state. Furthermore, the corresponding robust defense (RD) mechanisms are proposed to actively modify the observations based on the distances of sequential states. Experiments are conducted based on the standard IEEE reliability test system, and the results show that security risks indeed exist in the state-of-the-art DRL-based power system emergency control models. The effectiveness, stealthiness, instantaneity, and transferability of the proposed attacks and defense mechanisms are demonstrated with both white-box and black-box settings.