Future-proof energy systems with ML: tackling data drift and low-data forecasting.

An energizing meetup at Dexter Energy that’s for sure! First Aletta Heemskerk from Dexter Energy took the stage and enlightened us about her Bayesian approach for forecasting wind power generation, and specially the key part - fitting the wind speed-energy curve 🌬️📈

Key takeaways?
- The power of prior information allows for more accurate wind power predictions 🎯
- Bayesian modeling unlocks more transparency compared to parametric approaches as everything is viewed as a distribution 🔮
- Even sparse data can be handled, amongst others through hierarchical modeling 💡

Our second speaker Tatjana Puskarov presented about the methodologies and challenges they faced at Sensorfact when developing data drift detection for their ML system that prevents energy waste 🌱

What to account for when tackling data drift?
- Sounds like an open door, but, don’t drift away from all forms of drift; know what kind of drift you want to detect 🧐
- If you detect model drift check the detect and react framework; measure dissimilarity, test it and adapt❗️
- Your underlying data generating process is not stationary ❌

💙⚡️Big time credits to Dexter Energy for hosting us, the speakers Aletta Heemskerk & Tatjana Puskarov and all other co-organisers Andrei Kurylionak Alexander Backus, Andi Piftor, Lieke Kools, Louis de Bruijn & Alexander Oude Elferink ⚡️🧡