About

Inceptive’s predictive platform showcase, SPoC (Swiss Proof of Concept) is our technological demonstrator of our AI model delivery platform, Inceptive Server, in the context of energy.

This showcase uses data retrieved from the ENTSOE transparency platform of cross zonal energy transfers (CZET) and generation data of Switzerland. Then for each variable, a forecast with a basis from one hour to 24 hours are done.

The data is present on two different dashboards, one for generation variables (Hydro Pumped Storage, Hydro Water Reservoir, Solar, Nuclear and wind onshore) and another for CZET (with France, Italy, Austria and Germany/Luxembourg)

Requesting access

To request access, please fill the contact form and let us know your interest in the platform.

Platform content

The platform simply contains two dashboards for both generation and CZET. After successfully login, click on “View” on one of the two projects, then to Dashboard on the top navigation bar.

There are two filters :

• Date filter : Allowing to browse the length of all the graphics
• Prediction Horizon : Allowing the add more curves, each with the selected “lag” (e.g. time horizon). A lag of 12 will be the predictions done at hour H for the hour H+12. A lag of 0 means the real data retrieved from ENTSOE transparency platform.

About the accuracy of the models

Our models are simple and show the complexity and non-triviality of both problems.

We have created two models using Inceptive’s Igloo tools, one for the generation and one for CZET.

These model only use date time and the data of generation or CZET from the last seven days, wich “explains” an important part of the changes, but is clearly not enough. On one side for the generation of solar electricity, forecast is highly correlated with real outcomes. That’s because the information of day/night is enough for the model to understand the variability. But on the other side, nuclear generation is more related to other external factors not present in these data. So production changes are less highly anticipated by the model.

For us, these models show the interest of Machine Learning for forecasting these variables. But at the same time, the complexity and the need of data aggregation from different sources to provide solid forecasts.