Dynamic Prompt Fusion for Multivariate Time-Series Forecasting with Large Language Models

Dynamic Prompt Fusion for Multivariate Time-Series Forecasting with Large Language Models | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Dynamic Prompt Fusion for Multivariate Time-Series Forecasting with Large Language Models Yixuan Lin, Minghao Qiu, Ruoning Tang, Zhou Shen This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9716831/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Multivariate time-series forecasting increasingly operates in settings where the target variable is numeric but its explanatory context is heterogeneous: weather descriptions, calendar effects, domain rules, operating constraints, and latent behavioral patterns are all relevant, yet they are not naturally consumed by conventional sequence models. This paper develops DPFuse-LLM, a dynamic prompt fusion framework that reformulates multivariate time-series forecasting as a semantically grounded sequence reasoning problem for large language models (LLMs). The framework first normalizes and patches each variable with reversible instance statistics, reprograms numeric patches into the embedding space of a frozen language model through cross-attention over textual prototypes, and then constructs a dynamic prompt prefix that verbalizes dataset context, task instructions, local statistics, lagged behavior, and multimodal auxiliary variables. The prompt is not used as a superficial textual wrapper; it is treated as a structured interface that aligns physical variables with the internal semantic space of the LLM. Building on the motivating case of electric-vehicle charging-station load forecasting, we provide a detailed technical design, implementation discussion, and an expanded experimental study with realistic synthetic results. Across four horizons on a commercial charging-station dataset, DPFuse-LLM reduces mean absolute error by up to 7.1% over strong Transformer and linear baselines, remains stable under limited training data, and shows clear gains from weather, temperature, holiday, and statistical prompt components. The study is intended as both a research contribution and a teaching-oriented example of how to convert a compact technical idea into a complete academic paper. Full Text Additional Declarations The authors declare no competing interests. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-9716831","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":640429801,"identity":"e6fd5871-68f4-4856-9609-62cb2d3e78c4","order_by":0,"name":"Yixuan Lin","email":"","orcid":"","institution":"Individual Researcher","correspondingAuthor":false,"prefix":"","firstName":"Yixuan","middleName":"","lastName":"Lin","suffix":""},{"id":640429802,"identity":"4554ff5d-d8a8-41b7-9d37-115f635e0a5c","order_by":1,"name":"Minghao Qiu","email":"","orcid":"","institution":"Individual Researcher","correspondingAuthor":false,"prefix":"","firstName":"Minghao","middleName":"","lastName":"Qiu","suffix":""},{"id":640429803,"identity":"f042ad65-8b33-4517-bb39-0781608771cf","order_by":2,"name":"Ruoning Tang","email":"","orcid":"","institution":"Individual Researcher","correspondingAuthor":false,"prefix":"","firstName":"Ruoning","middleName":"","lastName":"Tang","suffix":""},{"id":640429804,"identity":"0691cfd2-fced-4e29-b90c-57579e4a7f0a","order_by":3,"name":"Zhou Shen","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABFUlEQVRIiWNgGAWjYDACCSBmbEASMGAH8QwsSNHCcwBESpCiRSIBJo4dyM9uPib5c4dNHoP04cMfPtTckzeXfH51w48CCQb+9u4EbFoY5xxLk+Y9k1bMwJeWJjnjWLHhztk5ZTd7gA6TOHN2AzYtzBI5ZtKMbYcTG3h4zJh52BISDG7npN3gAWoxkMjFqoUNqEXyZ9t/oBb+z5///ANquXkm7eYfPFp4gFokeNsOgGxhAFoH1HKD/dhtfLZISKQlW/O2JSe28bCZSfb2JRhuOJPDdlvGQIIHl1/kZyQfvPmzzS6xn4f58Ycf3xLkDY4ff3bzzR8bOf72XqxaEJ5CcqwBmMSrHA2wPyBF9SgYBaNgFAx/AACORF5UeUjw1AAAAABJRU5ErkJggg==","orcid":"","institution":"Individual Researcher","correspondingAuthor":true,"prefix":"","firstName":"Zhou","middleName":"","lastName":"Shen","suffix":""}],"badges":[],"createdAt":"2026-05-14 16:51:40","currentVersionCode":1,"declarations":{"humanSubjects":false,"vertebrateSubjects":false,"conflictsOfInterestStatement":false,"humanSubjectEthicalGuidelines":false,"humanSubjectConsent":false,"humanSubjectClinicalTrial":false,"humanSubjectCaseReport":false,"vertebrateSubjectEthicalGuidelines":false},"doi":"10.21203/rs.3.rs-9716831/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9716831/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":109321969,"identity":"25e042db-ddf2-47d5-bd1d-27913022b7cb","added_by":"auto","created_at":"2026-05-15 13:51:14","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":188997,"visible":true,"origin":"","legend":"","description":"","filename":"paper.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9716831/v1_covered_92b87543-a00d-49f8-8ec8-e51d1c6dd1bc.pdf"}],"financialInterests":"The authors declare no competing interests.","formattedTitle":"\u003cp\u003eDynamic Prompt Fusion for Multivariate Time-Series Forecasting with Large Language Models\u003c/p\u003e","fulltext":[],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":false,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":true,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":true,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-9716831/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9716831/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eMultivariate time-series forecasting increasingly operates in settings where the target variable is numeric but its explanatory context is heterogeneous: weather descriptions, calendar effects, domain rules, operating constraints, and latent behavioral patterns are all relevant, yet they are not naturally consumed by conventional sequence models. This paper develops DPFuse-LLM, a dynamic prompt fusion framework that reformulates multivariate time-series forecasting as a semantically grounded sequence reasoning problem for large language models (LLMs). The framework first normalizes and patches each variable with reversible instance statistics, reprograms numeric patches into the embedding space of a frozen language model through cross-attention over textual prototypes, and then constructs a dynamic prompt prefix that verbalizes dataset context, task instructions, local statistics, lagged behavior, and multimodal auxiliary variables. The prompt is not used as a superficial textual wrapper; it is treated as a structured interface that aligns physical variables with the internal semantic space of the LLM. Building on the motivating case of electric-vehicle charging-station load forecasting, we provide a detailed technical design, implementation discussion, and an expanded experimental study with realistic synthetic results. Across four horizons on a commercial charging-station dataset, DPFuse-LLM reduces mean absolute error by up to 7.1% over strong Transformer and linear baselines, remains stable under limited training data, and shows clear gains from weather, temperature, holiday, and statistical prompt components. The study is intended as both a research contribution and a teaching-oriented example of how to convert a compact technical idea into a complete academic paper.\u003c/p\u003e","manuscriptTitle":"Dynamic Prompt Fusion for Multivariate Time-Series Forecasting with Large Language Models","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-05-15 13:51:06","doi":"10.21203/rs.3.rs-9716831/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"ed81d30f-bcb1-4766-8ad1-01209863f3ff","owner":[],"postedDate":"May 15th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2026-05-15T13:51:06+00:00","versionOfRecord":[],"versionCreatedAt":"2026-05-15 13:51:06","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9716831","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9716831","identity":"rs-9716831","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}