From October 20 to 23, 2025, the ICCS-NTUA team successfully represented the Opentunity project at the IEEE Innovative Smart Grid Technologies (ISGT) Europe 2025 conference in Malta. Organized by IEEE PES, FIR.MT, and the IEEE Malta section, this prestigious forum brought together leading experts from the scientific community and the energy industry to explore the latest innovations in smart grid reliability.

Addressing Critical Infrastructure Risks

Power transformers are the backbone of the grid, but their failure can lead to cascading outages and significant financial losses.

Traditional monitoring methods are often reactive or rely on expensive,

 specialized sensors. To address this, the research presented by ICCS-NTUA, titled “Data Driven Model Agnostic Methodology for Transformers Top Oil Anomaly Detection,” proposes a proactive, predictive maintenance approach.

The core of the study is a methodology that detects anomalies in top-oil temperature, a primary indicator of a transformer’s health. Unlike complex physical models, this approach is “model-agnostic,” meaning it can be applied across different types of equipment using only minimal, standard measurements: load current, ambient temperature, and top-oil temperature.

Innovative Methodology and Real-World Validation

The research combines Deep Neural Networks (DNN) with Statistical Process Control (SPC). The process involves:

• Predictive Modeling: A machine learning model is trained on historical data to predict the expected top-oil temperature under normal conditions.
• Anomaly Detection: By comparing real-time measurements against these predictions, the system uses SPC control charts to identify deviations that signify potential faults, such as cooling pump failures or insulation degradation.

The effectiveness of this tool was validated using data from two real-world autotransformers currently operating in the Greek Transmission System. The results showed a high predictive accuracy, with R² values exceeding 0.98. Crucially, the methodology successfully identified anomalies that correlated with significant deviations in bushing capacitance, a clear diagnostic indicator of a developing issue.

Impact and Reception

The presentation was highly valued by attendees, who particularly praised the quality and accessibility of the dataset used for the research. The technical session also sparked a productive debate regarding the algorithm’s ability to distinguish between noise and real-world operational anomalies, highlighting the relevance of this work for future industrial applications.

By utilizing data already available via standard SCADA systems, this Opentunity-funded research provides grid operators with a cost-effective early warning system, enabling them to make informed maintenance decisions before a critical failure occurs.

Read the Full Paper

For a detailed look at the technical implementation, mathematical models, and complete case study results, you can access the full publication here:

Download Link: Data Driven Model Agnostic Methodology for Transformers Top Oil Anomaly Detection