Why Your Forecasting Transformer Isn’t Working (And How To Fix It...
summary
Implements a Temporal Fusion Transformer via PyTorch Forecasting and PyTorch Lightning, training on hourly PV data from Hugging Face, site metadata (capacity, orientation) and 14 ECMWF IFS NWP variables fused by timestamp. Builds a 36-hour continuous window with zero nulls and a horizon column, classifies static vs time-varying features, runs LR finder, early stopping and gradient clipping 0.1. Predicts unseen sites 18 hours ahead with close fits; top drivers are solar radiation, low cloud cover — data pipeline quality proves more important than model tweaks.
AI-written summary · about
related talks
31mFoundational Models for Time Series Forecasting: are we there yet?
21mFoundational Models for Time Series Forecasting: are we there yet?
31mFoundational Time Series Models in Practice: The Future of Forecasting, or Just Hype?
19mForecasting for Everyone - Foundation Models and the Future of Time Series Prediction
33mFrontiers of Foundation Models for Time Series
31mKeeping up with the renewable energy transition
01h 16mPredicting Weather Patterns using Time Series ML
54mTransformes for Time Series: Is the New State of the Art (SOA) Approaching?