BRIDGE: Big Data-Powered Large Language Models for Real-Time Multi-Modal Decision Support in Industrial Time Series

BRIDGE: Big Data-Powered Large Language Models for Real-Time Multi-Modal Decision Support in Industrial Time Series | 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 BRIDGE: Big Data-Powered Large Language Models for Real-Time Multi-Modal Decision Support in Industrial Time Series Fulin Li, Jun Nie, Zhijing Li This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9352751/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 8 You are reading this latest preprint version Abstract Industrial Internet-of-Things (IIoT) environments continuously generatemassive, heterogeneous streams of sensor readings, operational logs, andalarm messages, demanding intelligent systems capable of real-time decisionsupport across forecasting, anomaly detection, and natural-languageexplanation. Existing time-series foundation models address these tasks inisolation and ignore the rich graph-structured topology inherent toindustrial sensor networks, while general-purpose large language models(LLMs) lack efficient mechanisms for processing high-frequency numericalstreams under strict latency constraints. We propose BRIDGE ( B ig data-driven R eal-time I ndustrialmulti-modal D ecision support via G raph- E nhancedmulti-modal transformer), a unified pre-trained framework integrating threecomplementary modules: (i) a Graph-enhanced Topology Encoder (GTE)that captures latent sensor-network dependencies via multi-head graphattention, enriching temporal representations with physical topologypriors; (ii) an Extended-LSTM Temporal Backbone (xLSTM-TB) thatleverages true recurrence for coherent long-horizon probabilisticforecasting, augmented by our proposed Contiguous Graph PatchMasking (CGPM)---a graph-aware pre-training strategy that jointly maskstopologically adjacent sensor patches to improve robustness under realisticsensor-outage conditions; and (iii) a Cross-modal LLMAlignment (CLA) module that reprograms heterogeneous industrial tokensinto a frozen LLM's representation space using fewer than \((0.3)\) M additionalparameters, enabling zero-shot natural-language decision reports withoutbackbone fine-tuning. Pre-trained on a curated corpus of \((78)\) millionheterogeneous industrial time-series samples spanning manufacturing, energy,transportation, and cloud-infrastructure domains, BRIDGE is evaluatedon the GiftEval-ZS and Chronos-ZS zero-shot forecasting benchmarks and fourindustrial anomaly-detection benchmarks (SMAP, MSL, SMD, SWaT).BRIDGE achieves state-of-the-art zero-shot CRPS of \((\mathbf{0.396})\) on GiftEval-ZS, surpassing TiRex by \((3.7%)\) and TimesFM 2.0 by \((13.7%)\) ,while simultaneously attaining an \((F_1)\) score of \((\mathbf{92.4%})\) on SMAPanomaly detection and producing interpretable decision reports withROUGE-L of \((0.74)\) ---all within an end-to-end inference latency of \((\mathbf{12})\) ms, making BRIDGE the first framework to jointly achieveindustrial-grade forecasting accuracy, topology-aware anomaly detection,and explainable real-time decision support. Large language models Time-series foundation models Multi-modal fusion Zero-shot forecasting Full Text Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Reviewers agreed at journal 18 May, 2026 Reviews received at journal 30 Apr, 2026 Reviewers agreed at journal 20 Apr, 2026 Reviewers agreed at journal 15 Apr, 2026 Reviewers invited by journal 13 Apr, 2026 Editor assigned by journal 08 Apr, 2026 Submission checks completed at journal 08 Apr, 2026 First submitted to journal 08 Apr, 2026 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-9352751","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":626032363,"identity":"9016ff16-4060-4a6c-9031-aa3ae8b14d72","order_by":0,"name":"Fulin Li","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA3klEQVRIie3PMQrCMBSA4RcC6RLsasngFZ4IIljaqyiBTB16hIqDS8E1x+gR1IedPIBgBydnRwcHq9DVdBTMDwkE3kcSAJ/vB0POru0OIEb71fWOcdKDcPyQcMhpbHOj3QTepC2ywih5P7DCSQIuVJ43SXUpUcW44xDQsXI8TCiLN101J5xk2AxAGnP+TsJaSSSNuwx1hjcOQzl1kPaWjtAMiRV9SRJZs1xDT8LnFmkRSiJWotHC9Zd0S+ySPykVwWrzeDzjJAyo/kre8XYti+4kXOMdSfsM+nw+35/2AqV9RX4mLe6uAAAAAElFTkSuQmCC","orcid":"","institution":"Guangdong University of Science and Technology","correspondingAuthor":true,"prefix":"","firstName":"Fulin","middleName":"","lastName":"Li","suffix":""},{"id":626032370,"identity":"bb37df66-ce1c-4efa-943a-4dd84e94a10f","order_by":1,"name":"Jun Nie","email":"","orcid":"","institution":"Guangdong University of Science and Technology","correspondingAuthor":false,"prefix":"","firstName":"Jun","middleName":"","lastName":"Nie","suffix":""},{"id":626032371,"identity":"200a4d74-8f1d-4b64-add0-1a2f418587fd","order_by":2,"name":"Zhijing Li","email":"","orcid":"","institution":"Guangdong University of Science and Technology","correspondingAuthor":false,"prefix":"","firstName":"Zhijing","middleName":"","lastName":"Li","suffix":""}],"badges":[],"createdAt":"2026-04-08 06:54:13","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-9352751/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9352751/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":107489738,"identity":"ef2700f1-9bcf-4632-a26c-820013bf0471","added_by":"auto","created_at":"2026-04-22 02:48:53","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2629454,"visible":true,"origin":"","legend":"","description":"","filename":"BRIDGEBigDataPoweredLargeLanguageModelsforRealTimeMultiModalDecisionSupportinIndustrialTimeSeries1.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9352751/v1_covered_b222a3e9-65b1-4e94-b9bf-a01ae90bb0f7.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"BRIDGE: Big Data-Powered Large Language Models for Real-Time Multi-Modal Decision Support in Industrial Time Series","fulltext":[],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":false,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"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":"journal-of-king-saud-university-computer-and-information-sciences","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"Learn more about [Journal of King Saud University Computer and Information Sciences](https://link.springer.com/journal/44443)","snPcode":"44443","submissionUrl":"https://submission.springernature.com/new-submission/44443/3","title":"Journal of King Saud University Computer and Information Sciences","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Springer Open","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Large language models, Time-series foundation models, Multi-modal fusion, Zero-shot forecasting","lastPublishedDoi":"10.21203/rs.3.rs-9352751/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9352751/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eIndustrial Internet-of-Things (IIoT) environments continuously generatemassive, heterogeneous streams of sensor readings, operational logs, andalarm messages, demanding intelligent systems capable of real-time decisionsupport across forecasting, anomaly detection, and natural-languageexplanation. Existing time-series foundation models address these tasks inisolation and ignore the rich graph-structured topology inherent toindustrial sensor networks, while general-purpose large language models(LLMs) lack efficient mechanisms for processing high-frequency numericalstreams under strict latency constraints. We propose \u003cb\u003eBRIDGE\u003c/b\u003e(\u003cb\u003eB\u003c/b\u003eig data-driven \u003cb\u003eR\u003c/b\u003eeal-time \u003cb\u003eI\u003c/b\u003endustrialmulti-modal \u003cb\u003eD\u003c/b\u003eecision support via \u003cb\u003eG\u003c/b\u003eraph-\u003cb\u003eE\u003c/b\u003enhancedmulti-modal transformer), a unified pre-trained framework integrating threecomplementary modules: (i) a \u003cem\u003eGraph-enhanced Topology Encoder\u003c/em\u003e (GTE)that captures latent sensor-network dependencies via multi-head graphattention, enriching temporal representations with physical topologypriors; (ii) an \u003cem\u003eExtended-LSTM Temporal Backbone\u003c/em\u003e (xLSTM-TB) thatleverages true recurrence for coherent long-horizon probabilisticforecasting, augmented by our proposed \u003cem\u003eContiguous Graph PatchMasking\u003c/em\u003e (CGPM)---a graph-aware pre-training strategy that jointly maskstopologically adjacent sensor patches to improve robustness under realisticsensor-outage conditions; and (iii) a \u003cem\u003eCross-modal LLMAlignment\u003c/em\u003e (CLA) module that reprograms heterogeneous industrial tokensinto a frozen LLM's representation space using fewer than \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\((0.3)\\)\u003c/span\u003e\u003c/span\u003eM additionalparameters, enabling zero-shot natural-language decision reports withoutbackbone fine-tuning. Pre-trained on a curated corpus of \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\((78)\\)\u003c/span\u003e\u003c/span\u003e millionheterogeneous industrial time-series samples spanning manufacturing, energy,transportation, and cloud-infrastructure domains, BRIDGE is evaluatedon the GiftEval-ZS and Chronos-ZS zero-shot forecasting benchmarks and fourindustrial anomaly-detection benchmarks (SMAP, MSL, SMD, SWaT).BRIDGE achieves state-of-the-art zero-shot CRPS of \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\((\\mathbf{0.396})\\)\u003c/span\u003e\u003c/span\u003eon GiftEval-ZS, surpassing TiRex by \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\((3.7%)\\)\u003c/span\u003e\u003c/span\u003e and TimesFM 2.0 by \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\((13.7%)\\)\u003c/span\u003e\u003c/span\u003e,while simultaneously attaining an \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\((F_1)\\)\u003c/span\u003e\u003c/span\u003e score of \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\((\\mathbf{92.4%})\\)\u003c/span\u003e\u003c/span\u003e on SMAPanomaly detection and producing interpretable decision reports withROUGE-L of \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\((0.74)\\)\u003c/span\u003e\u003c/span\u003e---all within an end-to-end inference latency of\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\((\\mathbf{12})\\)\u003c/span\u003e\u003c/span\u003e ms, making BRIDGE the first framework to jointly achieveindustrial-grade forecasting accuracy, topology-aware anomaly detection,and explainable real-time decision support.\u003c/p\u003e","manuscriptTitle":"BRIDGE: Big Data-Powered Large Language Models for Real-Time Multi-Modal Decision Support in Industrial Time Series","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-04-21 12:13:33","doi":"10.21203/rs.3.rs-9352751/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"reviewerAgreed","content":"151127803450256212057442473054010396152","date":"2026-05-19T02:25:24+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-04-30T06:18:17+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"268085510873004168682405758686175701856","date":"2026-04-20T06:37:50+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"115443448534309208458873034568179727873","date":"2026-04-15T12:48:41+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-04-13T19:30:42+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-04-09T02:54:39+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-04-09T02:54:29+00:00","index":"","fulltext":""},{"type":"submitted","content":"Journal of King Saud University Computer and Information Sciences","date":"2026-04-08T06:47:35+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"journal-of-king-saud-university-computer-and-information-sciences","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"Learn more about [Journal of King Saud University Computer and Information Sciences](https://link.springer.com/journal/44443)","snPcode":"44443","submissionUrl":"https://submission.springernature.com/new-submission/44443/3","title":"Journal of King Saud University Computer and Information Sciences","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Springer Open","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"541abadb-ad0c-4b54-9e5e-265473a48729","owner":[],"postedDate":"April 21st, 2026","published":true,"recentEditorialEvents":[{"type":"reviewerAgreed","content":"151127803450256212057442473054010396152","date":"2026-05-19T02:25:24+00:00","index":16,"fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-04-30T06:18:17+00:00","index":13,"fulltext":""}],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-04-21T12:13:33+00:00","versionOfRecord":[],"versionCreatedAt":"2026-04-21 12:13:33","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9352751","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9352751","identity":"rs-9352751","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}