2 - Decoding the Sun

Background:

Switzerland is rapidly expanding its use of solar power. Driven by the "Energy Strategy 2050", the country has seen a massive increase in installed PV capacity, a trend that is set to accelerate. While this is great for clean energy, the unpredictable nature of PV generation (e.g., sudden changes in weather) creates significant challenges in matching electricity supply with demand in real-time. This mismatch is known as imbalance. Without reliable forecasts of the imbalance, Swissgrid would be forced to react to imbalances while they occur, a less efficient and more costly approach that could also jeopardize the security of the power supply.

Challenge:

Build a machine learning model that predicts the system imbalance 15 minutes ahead. Accurate forecasts allow Swissgrid to stabilize the grid at lower cost.

Data:

System imbalance: time series of measured values. PV forecasts: multiple updates for the same delivery time (forecast evolution).

What you will learn:

Python and time series forecasting. Machine learning (LSTM, transformers, etc.). Using PV forecast evolution as features.

A ready-to-use pipeline is provided; you may adapt it or develop your own. Submissions are scored by Mean Absolute Error (MAE).