Digitala Vetenskapliga Arkivet

Change search
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
Anomaly Detection for Multivariate Time Series Using Self-Organizing Map-Based Graph Neural Networks
KTH, School of Electrical Engineering and Computer Science (EECS).
2024 (English)Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesisAlternative title
Anomaly Detection för Multivariat tidsserie med hjälp av Självorganiserande kartbaserad graf Neurala nätverk (Swedish)
Abstract [en]

Detecting anomalies in operational data is crucial for effectively monitoring system performance and ensuring operational safety. However, the complexity and variability of operational data, particularly in anomaly detection within multivariate time series data, present a continuous challenge due to the need for analysis across both temporal and spatial dimensions. Various AI-driven methods, including Recurrent Neural Networks (RNNs) and Convolutional Neural Networks (CNNs), are frequently used to process and analyze time series data. While these models demonstrate promising performance, they often struggle with interpretability due to their complex architectures and the implicit analysis of spatial and temporal relationships. Graph Neural Networks (GNNs) have emerged as promising tools for explicitly analyzing complex dependencies and relationships in data by aggregating the neighbors and their relationships, but they encounter difficulties when working with unstructured inputs. To overcome this challenge, this thesis presents an approach that integrates Self-Organizing Maps (SOM) with GNNs. By converting unstructured time series data into graph-like structures, SOM enhances GNN’s capability to distinguish between normal and anomalous patterns. The model was evaluated on multiple public datasets, and the results indicate that the proposed approach outperforms existing benchmark methods, achieving the highest F1-score compared to other GNN-based models.

Abstract [sv]

Att upptäcka anomalier i driftsdata är avgörande för att effektivt övervaka systemets prestanda och säkerställa driftsäkerhet. Komplexiteten och variabi- liteten hos operativa data, särskilt vid avvikelsedetektering inom multivariata tidsseriedata, utgör dock en kontinuerlig utmaning på grund av behovet av analys över både tidsmässiga och rumsliga dimensioner. Olika AI-drivna metoder, inklusive Recurrent Neural Networks (RNN) och Convolutional Neural Networks (CNNs), används ofta för att bearbeta och analysera tidsseriedata. Även om dessa modeller visar lovande prestanda, kämpar de ofta med tolkningsbarhet på grund av deras komplexa arkitekturer och den implicita analysen av rumsliga och tidsmässiga relationer. Graph Neural Networks (GNN) har dykt upp som lovande verktyg för att explicit analysera komplexa beroenden och relationer i data genom att aggregera grannarna och deras relationer, men de stöter på svårigheter när de arbetar med ostrukturerade indata. För att övervinna denna utmaning presenterar denna avhandling ett tillvägagångssätt som integrerar Self-Organizing Maps (SOM) med GNN. Genom att konvertera ostrukturerade tidsseriedata till grafliknande strukturer förbättrar SOM GNN:s förmåga att skilja mellan normala och avvikande mönster. Modellen utvärderades på flera offentliga datauppsättningar, och resultaten indikerar att den föreslagna metoden överträffar befintliga benchmarkmetoder och uppnår det högsta F1-poängen jämfört med andra GNN-baserade modeller.

Place, publisher, year, edition, pages
2024. , p. 57
Series
TRITA-EECS-EX ; 2024:948
Keywords [en]
Graph Neural Network, Self-Organizing Map, Anomaly Detection, Multivariate Time Series
Keywords [sv]
Graph Neural Network, Självorganiserande karta, Anomali Detection, Mul- tivariate Time Series Graph Neural Network, Självorganiserande karta, Anomali Detection, Multivariate Time Series
National Category
Computer and Information Sciences
Identifiers
URN: urn:nbn:se:kth:diva-361103OAI: oai:DiVA.org:kth-361103DiVA, id: diva2:1943664
Supervisors
Examiners
Available from: 2025-03-17 Created: 2025-03-11 Last updated: 2025-03-17Bibliographically approved

Open Access in DiVA

fulltext(5386 kB)55 downloads
File information
File name FULLTEXT02.pdfFile size 5386 kBChecksum SHA-512
ab378049274739268c692211e8d0d9b456f69f1ca238788e54b7aa6c5f4856325d2ae532e9a090933948da7c7a0e862e498b15fbf2267027432a61dc967ed6c9
Type fulltextMimetype application/pdf

By organisation
School of Electrical Engineering and Computer Science (EECS)
Computer and Information Sciences

Search outside of DiVA

GoogleGoogle Scholar
Total: 55 downloads
The number of downloads is the sum of all downloads of full texts. It may include eg previous versions that are now no longer available

urn-nbn

Altmetric score

urn-nbn
Total: 311 hits
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf