My research centers on time series forecasting and explainability in digital twins. The work advances algorithms and system designs that make forecasts accurate, interpretable, and actionable in real time. My projects include an interpretable forecasting method that links predictions to characteristic historical patterns, enabling traceable, sample-level reasoning about outputs. Further, I have designed a model-agnostic post-hoc technique that provides multi-granular explanations down to the time-step level and reveals cross-channel dependencies in multivariate settings. I have also investigated decomposition-driven forecasting approaches to improve both predictive accuracy and interpretability across families of forecasting models. My research also proposes conceptual and architectural models for integrating explainable AI into digital twins, together with a real-time demonstration on an industrial manufacturing line. 

A selection from Stockholm University publication database

• Artificial intelligence in digital twins—A systematic literature review

  2024. Tim Kreuzer, Panagiotis Papapetrou, Jelena Zdravkovic. Data & Knowledge Engineering 151

  Article

  Artificial intelligence and digital twins have become more popular in recent years and have seen usage across different application domains for various scenarios. This study reviews the literature at the intersection of the two fields, where digital twins integrate an artificial intelligence component. We follow a systematic literature review approach, analyzing a total of 149 related studies. In the assessed literature, a variety of problems are approached with an artificial intelligence-integrated digital twin, demonstrating its applicability across different fields. Our findings indicate that there is a lack of in-depth modeling approaches regarding the digital twin, while many articles focus on the implementation and testing of the artificial intelligence component. The majority of publications do not demonstrate a virtual-to-physical connection between the digital twin and the real-world system. Further, only a small portion of studies base their digital twin on real-time data from a physical system, implementing a physical-to-virtual connection.

  Read more about Artificial intelligence in digital twins—A systematic literature review

• AI Explainability Methods in Digital Twins: A Model and a Use Case

  2025. Tim Kreuzer, Panagiotis Papapetrou, Jelena Zdravkovic. Enterprise Design, Operations, and Computing, 3-20

  Conference

  Digital twin systems can benefit from the integration of artificial intelligence (AI) algorithms for providing for example some predictive capabilities or supporting internal decision-making. As AI algorithms are often opaque, it becomes necessary to explain their decisions to a human operator working with the digital twin. In this study, we investigate the integration of explainable AI techniques with digital twins, which we termed XAI-DT system. We define the concept of XAI-DT system and provide a use case in smart buildings, where explainable AI is used to forecast CO₂ concentration. Further, we present a core architectural model for our digital twin, outlining its interaction with the smart building and its internal processing. We evaluate five AI algorithms and compare their explainability for the operator and the entire digital twin model based on standard explainability properties from the literature.

  Read more about AI Explainability Methods in Digital Twins

• A Meta-model for Integrating Explainable Forecasting with Digital Twins

  2025. Tim Kreuzer, Panagiotis Papapetrou, Jelena Zdravkovic. Advanced Information Systems Engineering Workshops, 169-180

  Conference

  Digital twins are virtual replicas of their physical counterparts, providing real-time monitoring and decision-making capabilities. By integrating forecasting-based methods, the potential of digital twins can be augmented significantly, enabling them to execute advanced predictive tasks. However, with digital twins typically involving a human-in-the-loop, the need for explainability becomes crucial for understanding how and why a forecast was made. To effectively integrate explainability methods, forecasting methods, and digital twins, it is essential to define the relations between these components in a structured manner. In this work, we address this issue by providing a meta-model for the integration of explainable forecasting methods with digital twins. We evaluate our meta-model in the context of a smart building digital twin with multiple forecasting and explainability methods. The evaluation demonstrates the inherent trade-off between providing explanations and generating accurate forecasts in this context.

  Read more about A Meta-model for Integrating Explainable Forecasting with Digital Twins

• Unpacking the trend: decomposition as a catalyst to enhance time series forecasting models

  2025. Tim Kreuzer, Jelena Zdravkovic, Panagiotis Papapetrou. Data mining and knowledge discovery 39 (5)

  Article

  For the time series forecasting task, several state-of-the-art algorithms employ moving-average decomposition for improved accuracy. However, the potential of decomposition techniques to enhance time series forecasting methods has not been explored in detail. In this work, we comprehensively investigate the use of decomposition methods for the forecasting task, comparing different decomposition techniques and their effect on forecasting accuracy, as well as the possibility of providing model-agnostic interpretability. We rework recent forecasting models to be compatible with any decomposition technique and experimentally evaluate their effectiveness in different forecasting setups. We further propose and assess a model-agnostic framework using decomposition for interpretability. Our results show that decomposition can improve forecasting accuracy, especially for the proposed decomposition-adapted models. Additionally, we demonstrate that the architectural choices of existing forecasting models can be improved by using different decomposition blocks internally. We found that decomposition techniques must be configured with a low number of components to provide model-agnostic interpretability. Our work concludes that decomposition can enhance time series forecasting algorithms, improving both their performance and interpretability.

  Read more about Unpacking the trend

Show all publications by Tim Kreuzer at Stockholm University