Country selection and coordination for PVNet training

[Sharkyii]

Country selection & coordination for PVNet training

This issue is to coordinate contributors who want to train PVNet models for new countries.

👉 Please comment here to:

• Say which country you’re working on
• Mention which step you’re starting with (PV data, NWP, configs, training, etc.)
• Tag others if you’re collaborating or handing over work

This helps avoid duplicate effort and keeps all discussion in one place.

Suggested countries to start with

• Netherlands
• Belgium
• Germany
• France

(Other countries are welcome if PV generation data is available and shareable.)

Reference guide

Use this doc as the main reference for the full workflow:
Document

Important notes

• Please don’t try everything in one PR — open PRs step-by-step
• Make sure pipelines run without errors before requesting dataset uploads
• Share training results (WANDB plots) in GitHub when ready

If you need help, tag @siddharth7113 or @peterdudfield.

[CodeVishal-17]

Hi! I’m interested in contributing to PVNet training.

I’d like to start with Netherlands and begin from:

• PV data exploration & availability check
• Initial config setup and pipeline validation

I’ll follow the reference workflow doc and open PRs step-by-step.
Happy to coordinate if someone is already working on this country.

[Sharkyii]

@peterdudfield Could you please review #124 when you get a chance?

I’ve added scripts for downloading the required data (Germany PV generation and GFS meteorological inputs). As a next step, I’ll verify full compatibility with the PVNet data pipeline and generation Zarr schema, and then upload the processed dataset to an S3 bucket for reproducible access.

[Her0n24]

Hi Maintainers and Contributors, nice to e-meet you.

I would like to work on the project with France data and I'm looking at the 15 minutes interval PV aggregates data by region from RTE France.

So far the guide in this repo is clear and good for beginners like me to follow, nevertheless I'm actively looking for someone to collaborate!

[Raakshass]

Hi @Sharkyii @peterdudfield @siddharth7113!

I'd like to contribute to PVNet training for India 🇮🇳

My Plan

Country: India
Starting with: PV data exploration & data source validation

Data Sources I've Identified

Source	Type	Resolution	Notes
Mendeley Dataset	Solar/Wind Generation	1-hour	Sep 2021 - Jun 2025, from Grid-India NERLDC
Kaggle Solar Plant Data	PV Generation	15-min	34 days, 2 plants with inverter-level data
National Power Portal (NPP)	Grid Generation	Daily/Real-time	Official POSOCO/Grid-India data
Ember India Data	Aggregated Solar	Monthly	State-level, API available

Proposed Steps

1. Week 1: Download and validate Mendeley hourly dataset
2. Week 2: Set up GFS NWP data pipeline for India region
3. Week 3: Create India configs aligned with existing PVNet structure
4. Week 4: Initial pipeline testing and sample generation

Questions

1. Is there a preferred minimum timespan for training data?
2. Should I focus on national-level or regional/state-level data initially?

Happy to collaborate with @Her0n24 (France) if there are shared pipeline components!

Will open PRs step-by-step as per the guidelines.

[Raakshass]

Hi! I've submitted a PR for the India contribution: #128

What's included:

• India solar data processing scripts (download, process, test, train)
• PVNet configuration for India solar + GFS NWP
• India regional grid metadata (5 regions: NR, WR, SR, ER, NER)
• Processed Mendeley dataset (Jan 2024 - Jun 2025) to Zarr format
• Baseline model achieving RMSE of 8,270 MW

Data source: Mendeley DOI 10.17632/y58jknpgs8.2

All CI checks pass. Looking forward to feedback!

[AswaniSahoo]

Hi @Sharkyii @peterdudfield @siddharth7113,

I'd like to take on Belgium 🇧🇪

Data Source: Elia (Belgian TSO) provides open solar generation data via
their Open Data Platform: https://opendata.elia.be/

Specifically, the "Solar power generation" dataset provides:

• 15-minute resolution measured & monitored solar capacity
• Historical data available from 2012+
• RESTful API + CSV export

My plan (step-by-step PRs per issue guidelines):

PR 1: Data collection script — fetch Elia solar generation via API,
output to Zarr format compatible with ocf-data-sampler
(per @siddharth7113's feedback on PR #127)

PR 2: Belgium-specific PVNet config + GFS NWP coordinate setup

PR 3: Pipeline validation + training run

Starting with PR 1 this week. Will share progress here.