In December 2020, enercast described the TRANSFER research project with Fraunhofer IEE, the University of Kassel's Intelligent Embedded Systems group and transmission system operator 50 Hertz. The research addressed a practical problem: new or modified energy-system components may have little historical data of their own.
Learning when site history is limited
A model trained on measured production needs observations that connect input conditions to output. A new plant cannot provide a long operating history immediately. Transfer learning explores whether knowledge from comparable components can help establish a model for a new one. Similarity must be assessed rather than assumed: equipment, location and operating behaviour can differ.
What the project proposed
The report described using metadata to compare source and target components and linking an appropriate existing model with information from the new component. Its examples included new wind and PV feed-in, PV with storage, and transformer power flows. The goal was automated knowledge transfer that could be tested against established approaches using real cases and datasets.
Research aims are not guaranteed results
A transferred model can be inappropriate when the new component differs from its source. Evaluation must examine such negative transfer and update the model as reliable local measurements become available. This article preserves the 2020 research announcement without claiming current project status, a completed universal model or a fixed accuracy improvement for new plants.