论文列表

Supply chain planning is a core challenge in supply chain management, orchestrating heterogeneous item categories across multiple vendors, and jointly optimizing end-to-end material flows, from sourcing and inbound logistics, through warehousing, to on-time delivery at demand locations. Electric power enterprises, as large and safety-critical consumers of engineering materials, require a comprehensive supply chain planning system, spanning demand planning, supplier and contract management, inventory control, to ensure reliable and cost-effective material availability. Moreover, recent advances in artificial intelligence (AI), particularly large language models (LLMs), have opened new directions for supply chain management, enabling knowledge-grounded reasoning, tool-augmented integration with enterprise platforms, and multi-agent coordination for closed-loop, adaptive planning. Accordingly, we present GridSCAPE, an LLM-based multi-agent framework, built to orchestrate closed-loop, real-time, intelligent supply chain planning for electric power enterprises. GridSCAPE deploys coordinated LLM agents with tool support, covering the entire supply chain, from demand and sourcing to inventory, warehousing, and logistics. We present implementation details of GridSCAPE and validate its efficacy via a utility case study and controlled experiments, showing reduced end-to-end cycle time, higher plan accuracy, and an increased in-stock rate.

Hybrid clouds offer more capacity and flexibility than public or private clouds alone, making them popular in business settings. However, they also complicate the scheduling of workflow applications due to their complex computing resources. Furthermore, significant concerns about exposing sensitive data on public clouds and securing data transmission add additional challenges to the scheduling process. Although various workflow scheduling strategies for hybrid clouds that address privacy and security concerns have been proposed, they have inherent limitations. Many of these methods, including heuristic and metaheuristic approaches, are tailored for batch processing and are not suitable for real-time scenarios where workflows can arrive unpredictably. Additionally, they often prioritize security over reducing execution time and costs, compromising overall efficiency. To tackle these challenges, we introduce a scheduling system in Hybrid clOuds that considers Privacy and Security constraints, called HOPS. HOPS utilizes Deep Reinforcement Learning to dynamically assign workflows to virtual machines in real-time, and aims to minimize both makespan and operational costs while adhering to privacy and security standards. We provide a comprehensive overview of the design of HOPS, and our experimental results clearly show its advantages over existing methods in terms of makespan, cost efficiency, and the success rate, with privacy and security compliance.

With the proliferation of cloud computing and the escalating demand for extensive data processing capabilities, an increasing number of enterprises are embracing hybrid cloud solutions. However, as more businesses move toward hybrid clouds, the need for effective solutions to privacy and security concerns becomes increasingly important. Although current scheduling approaches for cloud computing have addressed privacy protection to some extent, few have adequately considered the unique challenges posed by hybrid clouds. To address this gap, we propose a novel approach for scheduling jobs in hybrid clouds that prioritizes privacy protection. Our approach, called PH-DRL, leverages Deep Reinforcement Learning (DRL) to intelligently allocate jobs to virtual machines, optimizing both privacy and Quality of Service (QoS), while minimizing response time. We present the detailed implementation of our approach and our experimental results demonstrate the superior performance of PH-DRL in terms of privacy protection compared to existing methods.

Scheduling workflows in hybrid cloud environments presents significant challenges due to the inherent complexity of workflows and the dynamic nature of cloud resources. This complexity is further increased when attempting to balance workflow performance with privacy protection. Recent efforts have leveraged deep reinforcement learning (DRL) to address these challenges. However, most of these approaches rely on single-agent models, which can lead to security issues and scalability problems due to their centralized processing. Specifically, the properties of workflows are transferred to the single agent, which risks leaking privacy information. Our paper addresses these issues by introducing MARS, a real-time workflow scheduling method that prioritizes privacy protection in hybrid clouds. MARS leverages multi-agent deep reinforcement learning (MADRL) to optimize the workflow scheduling of cloud virtual machines (VMs). The benefit of our solution is that it relies on the collaborative learning of multi-agents on multiple VMs, which could assign user data to specific cloud servers for privacy protection while sharing training experiences between agents. In our implementation, MARS aims to reduce workflow completion time and operational costs while complying with strict privacy protection guidelines. The experimental results demonstrate that MARS can significantly surpass existing methods, reducing makespan by an average of 53.18% and costs by 61.98% compared to basic techniques, and achieving 20.26% and 25.71% improvements over the latest advanced methods, respectively.

The rapid development of vehicular networks has led to widespread adoption of various electric vehicle (EV) applications, often necessitating the deployment of large-scale deep neural networks (DNNs). However, constrained computational and energy resources pose challenges for executing computationally intensive DNN tasks exclusively within EVs. To address this issue, one potential solution is to utilize edge or cloud computing resources for collaborative computation, typically implemented through DNN partitioning and task offloading. Therefore, we propose a novel approach in this paper, named TOCC, to execute EV-generated DNN tasks with edge-cloud collaboration. Specifically, we first construct a performance prediction model, which can accurately predict the performance of different layers in a DNN. Then, we define the joint optimization problem of minimizing processing delay and energy consumption as a Markov Decision Process (MDP). Finally, we employ Deep Reinforcement Learning (DRL) to design a strategy that enables EVs to make optimal decisions for DNN task partitioning and offloading. The experimental results demonstrate that TOCC surpasses existing approaches in terms of both processing delay and energy consumption, and is applicable to various DNN types.

The distributed architecture of cloud computing necessitates robust defense mechanisms to secure network-accessible resources against a diverse and dynamic threat landscape. A Network Intrusion Detection System (NIDS) is pivotal in this context, with its efficacy in cloud environments hinging on its adaptability to evolving threat vectors while mitigating false positives. In this paper, we present a novel NIDS algorithm, anchored in the Transformer model and finely tailored for cloud environments. Our algorithm melds the fundamental aspects of network intrusion detection with the sophisticated attention mechanism inherent to the Transformer model, facilitating a more insightful examination of the relationships between input features and diverse intrusion types, thereby bolstering detection accuracy. We provide a detailed design of our approach and have conducted a thorough comparative evaluation. Our experimental results demonstrate that the accuracy of our model is over 93%, which is comparable to that of the CNN-LSTM model, underscoring the effectiveness and viability of our Transformer-based intrusion detection algorithm in bolstering cloud security.

Cloud computing has gained popularity in recent years, but with its rise comes concerns about data security. Unauthorized access and attacks on cloud-based data, applications, and infrastructure are major challenges that must be addressed. While machine learning algorithms have improved intrusion detection systems in cloud data security, they often fail to consider the entire life cycle of file processing, making it difficult to detect certain issues, especially insider attacks. To address these limitations, this paper proposes a novel approach to analyzing data file processing in multi-cloud environments using process mining. By generating a complete file processing event log from a multi-cloud environment, the proposed approach enables detection from both control flow and performance perspectives, providing a deeper understanding of the underlying file processing in its full life cycle. Through our case study, we demonstrate the power and capabilities of process mining for file security detection and showcase its ability to provide further insights into file security in multi-cloud environments.