Skip to main content
All CollectionsResources & MethodologyPower
Kpler Short term forecasts methodology
Kpler Short term forecasts methodology

Wind, solar generation, demand and SPOT price forecasts up to 46 days ahead

Hamza Aourach avatar
Written by Hamza Aourach
Updated over a week ago

(S)FTP access

Once you receive your credentials, you can access our FTP through:

Host: calc1.cor-e.fr

(FTP) Port: 21

(SFTP) port: 2222

Coverage

Zone

Live since

Wind

Solar

Load

Price DA

Austria

2019

Belgium

2019

Bosnia and Herzegovina

2025

Bulgaria

2025

Croatia

2024

Czech Republic

2023

DK1

2025

DK2

2025

Estonia

2025

Finland

2025

France

2019

Georgia

2025

Germany

2019

Greece

2025

Hungary

2024

Ireland

2025

Italy

2021

IT-Calabria

2025

IT-Centre-Nort

2025

IT-Centre-South

2025

IT-North

2025

IT-Sardinia

2025

IT-Sicily

2025

IT-South

2025

Latvia

2025

Lithuania

2025

Moldova

2025

Montenegro

2025

Netherlands

2019

NO1

2025

NO2

2025

NO3

2025

NO4

2025

NO5

2025

Poland

2023

Portugal

2025

Romania

2024

SE1

2025

SE2

2025

SE3

2025

SE4

2025

Serbia

2025

Slovakia

2024

Slovenia

2024

Spain

2021

Switzerland

2021

United Kingdom

2020

XK

2025

Documentation of wind, solar and demand forecasts

Kpler short term models rely on two external weather forecast providers:

  • ECMWF (a.k.a. EC) - European Centre for Medium-Range Weather Forecasts

  • GFS - Global Forecast System of the National Oceanic and Atmospheric Administration

Each weather provider generates 4 weather runs per day (00/06/12/18) for 2 models. There is one deterministic model called OP and one ensemble providing several scenarios (30 for GFS and 50 for EC). The table below contains the update intervals of our forecasts.

Note that in addition to new weather runs, at Kpler we make additional updates for price only (cf UPD in the table) 4 times a day with the latest information available. These updates rewrite the file for the given run. Example for run 00:

  • a new file is generated between 6 and 7am UTC

  • it is then updated a first time at 10am CET

  • it receives a final update around 11.15am CET

Points in time seen from the latest updated are available on the FTP. The files are not rewritten after that point.

The timings on the table above are for the first forecasted hour, then it can take up to ~1h to generate a value for the last forecasted hour.

Access on the FTP

Paths for each dataset:

  • wind generation: 4_Supply/Forecast/Generation/COR_E/{Country}/yyyy/mm

  • solar generation: 4_Supply/Forecast/Generation/COR_E/{Country}/yyyy/mm

  • demand: 3_Demand/Forecast/Load/COR_E/{Country}/yyyy/mm

  • Spot price: 2_Price/Forecast/Day_Ahead/COR_E/{Country}/yyyy/mm

File naming convention:

  • wind & solar generation: CORE_SUPPLY_FORECAST_Generation_CORE_{Country}_{Wind|Solar}_Hourly_{EC|GFS}_{OP|ENS}_yyyymmdd{00|06|12|18}.csv

  • demand: CORE_DEMAND_FORECAST_Load_CORE_{Country}_Hourly_{EC|GFS}_{OP|ENS|ENS_SCENARIOS}_yyyymmdd{00|06|12|18}.csv

  • Spot price: CORE_PRICE_FORECAST_DayAhead_CORE_{Country}_Hourly_{EC|GFS}_{OP|ENS|ENS_SCENARIOS}_yyyymmdd{00|06|12|18}.csv

For Ensemble forecasts, two files are published for each run:

  • one {ENS} containing the control scenario "0"

  • one {ENS_SCENARIOS} containing the remaining scenarios (30 for GFS and 50 for EC)

Price modelling methodology

Our model is combining machine learning and time serie analysis to predict hourly prices based on specific market conditions (production, consumption, fuel prices, interconnector capacity).

This artificial intelligence method called machine learning is a method based on fundamental analysis that is effective in predicting different products based on the evolution of historical data and by combining several parameters. To predict the wholesale market price we use the following variables:

  • Annual shape and level of electricity demand

  • Installed capacities of renewable energies

  • Fossil and nuclear power plant capacities

  • Fuel (gas, coal, carbon) prices

  • Interconnector constraints

From the changes of these different variables, our model is able to predict for each of the future hours a balance between supply and demand on the merit order for the wholesale market and the bid curves for the reserve prices as it has been studied historically.

In order to avoid the “black box effect” our modeling is performed in two steps:

  • Conversion of raw weather data (°C, wind m/s, solar Watts of radiation, precipitations, cloud cover etc..) into forecasts of consumption, solar production, renewable production

  • Conversion of these MW into €/MWh (prices forecast)

Every country model is fed with variables from surrounding countries, but is built and runs independently. Our short term models run using GFS & EC weather data, both Operational and Ensemble scenarios. It gets updated and published every time a new full day of weather forecast is made available

Our models are retrained on a daily basis over an ever extending window going from 2015 to the most recent actual value.

Did this answer your question?