[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-82731-en":3,"doc-seo-82731-105":29,"detail-sidebar-cat-0-en-105":91},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":20,"is_deleted":4,"is_public":20,"is_downloadable":20,"audit_status":20,"page_count":21,"language":22,"language_code":23,"site_id":24,"html_lang":23,"table_of_contents":25,"faqs":26,"seo_title":13,"seo_description":14,"update_tm":27,"read_time":28},82731,4398048949847,"Eliana","https://ap-avatar.wpscdn.com/avatar/400002536579ef2da7f?_k=1778318612642679267",8,"Research & Report","Hierarchical Multi-Agent Reinforcement Learning for Carbon-Aware AI Data Centers in Power Distribution Systems","Eco-friendly energy management for artificial intelligence data centers (AIDCs) is essential as AI-driven growth escalates energy use and carbon emissions. The framework introduces hierarchical carbon-aware multi-agent reinforcement learning (CA-MARL) for robust, efficient operation under uncertainty, while keeping power distribution systems low-carbon. It combines a workload manager agent that spatially assigns training and inference across AIDCs, and local AIDC agents that schedule temporal shifting, GPU and inference allocation, and cooling supply-air temperature using nodal carbon intensity derived from carbon emission flow integration. Evaluated on an IEEE 33-node power distribution system.","Hierarchical Multi-Agent Reinforcement Learning for Carbon-Aware AI Data Centers in Power Distribution Systems  \nHyunsoo Lee, Panggah Prabawa, Graduate Student Member, IEEE, Dae-Hyun Choi, Member, IEEE, and  \nJoongheon Kim, Senior Member, IEEE  \narXiv :2607 .03324v2 [ ee ss . SY] 7 Jul 2026  \nAbstract—Eco-friendly energy management for artificial intelligence data centers (AIDCs) is crucial because of the significant increase in energy consumptioninduced carbon emissions from AIDCs resulting from the rapid expansion of AI applications. This paper proposes a hierarchical carbon-aware multi-agent reinforcement learning (CA-MARL) framework for robust and efficient operations of AIDCs under uncertainties while ensuring low-carbon operation of power distribution systems. The framework comprises a workload manager (WM) agent and multiple local AIDC agents trained using a multi-agent transformer method, corresponding to a global AIDC aggregator and a local AIDC operator, respectively. Leveraging AIDC operation data along with nodal carbon intensity (NCI) calculated from the carbon emission flow-integrated distribution system operator problem, the WM agent spatially allocates AI training and inference jobs among all AIDCs. Based on the jobs allocated from the WM agent and NCI information, each AIDC agent schedules economical and eco-friendly operations of the AIDC by performing the following tasks: i) temporal shifting of training jobs, ii) spatial allocation of training graphics processing unit (GPU) blocks and inference GPUs within the AIDC, and iii) control of the supply air temperature of the cooling system. The effectiveness of the proposed framework was assessed using an IEEE 33-node power distribution system.  \nIndex Terms—AI data centers, carbon emission flow, multi-agent reinforcement learning, workload scheduling, power distribution systems.  \nI. Introduction  \nTHE rapid proliferation of large language models  \n(LLMs) and generative artificial intelligence (AI) has significantly increased the energy demand of AI data centers (AIDCs) . For AIDCs, training a single large-scale model can require several MWh of electricity [ 1] . Moreover, heterogeneous global-scale inference workloads maintain a persistent baseline energy consumption [2]. By 2024, AIDCs accounted for approximately 1 .5% of global electricity use, with expectations of doubling this percentage  \nH. Lee is with LG Electronics, 17709, Pyeongtaek, Gyeonggi-do, Republic of Korea (e-mail: [hyunsoo11.lee@lge.com](hyunsoo11.lee@lge.com)).  \nJ. Kim is with the Department of Electrical and Computer Engineering, Korea University, Seoul 02841, Republic of Korea (e-mail: [joongheon@korea.ac.kr](joongheon@korea.ac.kr)).  \nP. Prabawa and D.-H. Choi are with the School of Electrical and Electronics Engineering, Chung-Ang University, Seoul 06974, Republic of Korea (e-mails: {panggah22, [dhchoi](dhchoi}@cau.ac.kr)[}](dhchoi}@cau.ac.kr)[@cau.ac.kr](dhchoi}@cau.ac.kr)).  \nD.-H. Choi and J. Kim are the corresponding authors of this paper.  \nby 2030 [3] . This surging energy consumption of AIDCs generates a substantial carbon footprint and may lead to operational instability in the power distribution grids. Therefore, efficient and eco-friendly energy management of power distribution grids and AIDCs, utilizing distribution system operators (DSOs) and AIDC operators, respectively, is becoming increasingly critical.  \nCompared with conventional Internet data centers (IDCs), AIDCs are characterized by GPU-centric computing environments, higher power density, and AI training workloads that provide greater temporal and spatial scheduling flexibility. Unlike the IDCs, which are independent of AI applications, AIDCs manage two types of AI workloads with different characteristics: training and inference workloads [4]. The former refers to delay-tolerant workloads (e.g., deep learning training) that can tolerate relaxed service delays. For these workloads, temporal shifting is defined as a sche","cbCaioOtQxZ0Cr0E","https://ap.wps.com/l/cbCaioOtQxZ0Cr0E","pdf",2996881,1,16,"English","en",105,"# Introduction\n## Background: energy and carbon challenges of AIDCs\n## Training vs. inference workload scheduling\n## Need for joint DSO–AIDC low-carbon coordination","[{\"question\":\"What problem does the paper address for AI data centers?\",\"answer\":\"It targets eco-friendly energy management for AI data centers, focusing on reducing carbon emissions driven by increased energy consumption and maintaining stability in power distribution grids.\"},{\"question\":\"How does the proposed hierarchical CA-MARL framework coordinate decisions?\",\"answer\":\"A workload manager agent spatially allocates training and inference jobs across multiple AIDCs, while local AIDC agents schedule economical and eco-friendly operations using carbon-aware information.\"},{\"question\":\"Which operational controls are optimized at each AIDC agent?\",\"answer\":\"Each AIDC agent optimizes temporal shifting of training jobs, spatial allocation of GPU blocks and inference GPUs, and control of the cooling system’s supply air temperature.\"}]",1784182550,40,{"code":4,"msg":30,"data":31},"ok",{"site_id":24,"language":23,"slug":32,"title":13,"keywords":33,"description":14,"schema_data":34,"social_meta":86,"head_meta":88,"extra_data":90,"updated_unix":27},"hierarchical-multi-agent-reinforcement-learning-for-carbon-aware-ai-data-centers-in-power-distribution-systems","",{"@graph":35,"@context":85},[36,53,68],{"@type":37,"itemListElement":38},"BreadcrumbList",[39,43,47,50],{"item":40,"name":41,"@type":42,"position":20},"https://docshare.wps.com","Home","ListItem",{"item":44,"name":45,"@type":42,"position":46},"https://docshare.wps.com/document/","Document",2,{"item":48,"name":12,"@type":42,"position":49},"https://docshare.wps.com/document/research-report/",3,{"item":51,"name":13,"@type":42,"position":52},"https://docshare.wps.com/document/hierarchical-multi-agent-reinforcement-learning-for-carbon-aware-ai-data-centers-in-power-distribution-systems/82731/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":23,"description":14,"dateModified":61,"datePublished":62,"encodingFormat":60,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":40,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-17","2026-07-16",true,{"@type":65,"interactionType":66,"userInteractionCount":20},"InteractionCounter",{"@type":67},"ViewAction",{"@type":69,"mainEntity":70},"FAQPage",[71,77,81],{"name":72,"@type":73,"acceptedAnswer":74},"What problem does the paper address for AI data centers?","Question",{"text":75,"@type":76},"It targets eco-friendly energy management for AI data centers, focusing on reducing carbon emissions driven by increased energy consumption and maintaining stability in power distribution grids.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How does the proposed hierarchical CA-MARL framework coordinate decisions?",{"text":80,"@type":76},"A workload manager agent spatially allocates training and inference jobs across multiple AIDCs, while local AIDC agents schedule economical and eco-friendly operations using carbon-aware information.",{"name":82,"@type":73,"acceptedAnswer":83},"Which operational controls are optimized at each AIDC agent?",{"text":84,"@type":76},"Each AIDC agent optimizes temporal shifting of training jobs, spatial allocation of GPU blocks and inference GPUs, and control of the cooling system’s supply air temperature.","https://schema.org",{"og:url":51,"og:type":87,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":89,"canonical":51},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":92},[93,97,101,105,110,115,119,122,127,130,134],{"id":20,"doc_module":4,"doc_module_name":45,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":46,"doc_module":4,"doc_module_name":45,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":45,"category_name":102,"show_sort_weight":103,"slug":104},"Exam",70,"exam",{"id":106,"doc_module":4,"doc_module_name":45,"category_name":107,"show_sort_weight":108,"slug":109},5,"Comic",60,"comic",{"id":111,"doc_module":4,"doc_module_name":45,"category_name":112,"show_sort_weight":113,"slug":114},6,"Technology",50,"technology",{"id":116,"doc_module":4,"doc_module_name":45,"category_name":117,"show_sort_weight":28,"slug":118},7,"Healthcare","healthcare",{"id":11,"doc_module":4,"doc_module_name":45,"category_name":12,"show_sort_weight":120,"slug":121},30,"research-report",{"id":123,"doc_module":4,"doc_module_name":45,"category_name":124,"show_sort_weight":125,"slug":126},9,"Religion & Spirituality",20,"religion-spirituality",{"id":125,"doc_module":4,"doc_module_name":45,"category_name":128,"show_sort_weight":125,"slug":129},"World Cup","world-cup",{"id":131,"doc_module":4,"doc_module_name":45,"category_name":132,"show_sort_weight":131,"slug":133},10,"Lifestyle","lifestyle",{"id":135,"doc_module":4,"doc_module_name":45,"category_name":136,"show_sort_weight":106,"slug":137},19,"General","general"]