Project Description

The availability of critical infrastructures is of fundamental importance to the economy and society. Given the complexity of increasingly networked technical systems, the early detection of security-relevant processes and critical operating situations, as well as the initiation of suitable measures, is an enormous challenge. While the established monitoring of critical system components via conventional sensors serves as a crucial tool for detecting security-related incidents, the integration of additional data sources, such as social media, can enhance situational awareness, facilitate timely decision-making, and strengthen resilience against various external threats and disruptions.

In ISAKI, a system for detecting anomalies for use in critical infrastructures is being developed and tested, as exemplified in the area of water supply. The intended technical solution should provide indications of anomalies by continuously comparing the current operating status of a system with the expected status. The system is not limited to the available sensor measurements of the critical infrastructure provider but also integrates relevant information from a variety of data sources. This includes not only natural disasters but also human-induced disruptions such as protests or cyberattacks, which may not be immediately detected by traditional sensor networks.

For this purpose, various data sources are fused together, ranging from monitoring devices, specific sensors, environmental data and situation information from security authorities. This diverse data is automatically analyzed using artificial intelligence methods. Results and indications of anomalies are processed, taking into account aspects of ergonomics and acceptance by the plant operators, in such a way that relevant incidents are reliably detected and appropriate measures are initiated.

Results

The ISAKI Project concluded 2023. Results of this project include the development of Deep Learning models for event detection, information type classification (Seeberger and Riedhammer, 2022; Seeberger et al., 2023), and event summarization (Seeberger and Riedhammer, 2022; Seeberger and Riedhammer, 2023). Methods related to this project were published at the following conferences:

  1. Seeberger, P., Riedhammer, K., 2023. Multi-Query Focused Disaster Summarization via Instruction-Based Prompting, in: Proc. of the 32nd Text REtrieval Conference (TREC 2023). National Institute of Standards and Technology (NIST).
  2. Seeberger, P., Bocklet, T., Riedhammer, K., 2023. Information Type Classification with Contrastive Task-Specialized Sentence Encoders, in: Proc. 19th Conference on Natural Language Processing (KONVENS 2023). Association for Computational Linguistics, pp. 180–186 .
  3. Seeberger, P., Riedhammer, K., 2022. Combining Deep Neural Reranking and Unsupervised Extraction for Multi-Query Focused Summarization, in: Proc. of the 31st Text REtrieval Conference (TREC 2022). National Institute of Standards and Technology (NIST).
  4. Seeberger, P., Riedhammer, K., 2022. Enhancing Crisis-Related Tweet Classification with Entity-Masked Language Modeling and Multi-Task Learning, in: Proc. 2nd Workshop on NLP for Positive Impact. Association for Computational Linguistics, pp. 70–78 .

Partners

The BMBF-funded project is comprised of an interdisciplinary team that integrates user needs, research, experience, and societal interests concerning supply security. Erlanger Stadtwerke AG (ESTW) is participating as a user, primarily focusing on water and energy supply. The start-up Traversals Analytics and Intelligence plays a significant role in developing the demonstrator and provides the platform for it. Additionally, expertise in security aspects is contributed by the police department (Mittelfranken) and DITS.center, incorporating their existing experience into the project.