1 - Modeling PV Production with Machine Learning

Challenge

For small photovoltaic systems (<30kWp), CKW currently only measures the energy fed into its grid. This creates a gap in understanding the actual energy production and self-consumption of these systems. The objective of this challenge is to bridge that gap by developing a machine learning model that can accurately estimate total energy production and self-consumption.

Develop a machine learning model to estimate the energy production of small photovoltaic (PV) systems (<30kWp) using feed-in data, weather information, and egid data. The goal is to enable innovative and intelligent grid- and market-oriented solar energy solutions.

Why it matters for CKW

This challenge has significant real-world implications:

• Starting next year, PV production will be limited to 70% of the panel’s peak output as per new regulations.
• Accurate production estimates will help:
  • Quantify energy losses due to these limits.
  • Support the creation of innovative, market-oriented solar energy products.
  • Enhance CKW’s ability to make solar energy more attractive and efficient for customers.

Prerequisites

• Renku or Local IDE like VSCode or Pycharm
• Python 3.10+
• Git
• pip

Getting Started

1. Clone repository

   git clone https://gitlab.com/edhd/2025/modeling-pv-production-with-machine-learning/team1.git

2. Install requirements

   pip install -r requirements.txt

Data

• CKW’s grid includes approximately 9,600 PV systems smaller than 30kWp without production measurement.
• Participants will work with:
  • Time series energy data
  • Weather data (provided by Meteomatics)
  • EGID data
• The dataset is preprocessed with basic feature engineering, allowing participants to focus on building the best possible model.
• Both traditional machine learning approaches and advanced neural networks, such as LSTMs, are encouraged.

Input data

The data is found on a azure blob storage and can be accessed directly in renku or with a connection string provided by the challenge owner at the Hackday.

Results and machine learning models

The results and created machine learning models can be found in the models folder

Group Members

• Enrique Romano (Challenge owner)
• First Name (role)
• First Name (role)

License

This project is licensed under the Apache License 2.0 - see the LICENSE file for details.