Context
As renewable energy grows, energy planners depend on wind-power forecasts for planning, storage and flexibility decisions. But a traditional point forecast — one number per hour — says nothing about how wrong it might be, and it is exactly that uncertainty that makes planning hard.
That became the real question of the thesis: not how do we predict wind power most accurately, but how do we give planners a forecast they can actually make risk-aware decisions on?
Discovery & framing
We framed the work around decision-making under uncertainty rather than model accuracy alone — treating the forecast as decision support for risk-aware planning in digital energy, with an eye on how such a capability could fit into energy platforms.
Data & method
- Real wind and weather measurements from Sweden (SE1, SE3)
- Station selection near wind parks
- u/v wind-vector features
- Cyclical time features
- End-to-end ML pipeline with preprocessing and evaluation
- Temporal Fusion Transformer (TFT) with quantile regression
Product / platform relevance
Uncertainty ranges instead of a single value let planners decide with a clearer view of risk. An AI output is only useful when its uncertainty is presented in a way that supports the user's actual decision.
- Demand-side flexibility and EV charging
- Smart-building and industrial load shifting
- Storage optimisation and load balancing
Images
No illustrative or simulated result curves are used — only real material from the thesis work.
Reflection