Skip to main content

Electric consumption & production forecast

Use our models and AI databases to industrialize your forecasting models by integrating your business constraints.

Case Study

Use weather forecasts to predict the consumption and production of electricity on the distribution system at different grid meshes in the short and long term.

Background

Our client wishes to strengthen its role as a data operator and is committed to a dynamic of data openness. The aim is to make available analyses such as changes in consumption and production connected to the public distribution network managed by our client or information on the means implemented and the results obtained. In order to integrate renewable energy production into the distribution network, many of our client's internal activities require increasingly efficient forecasting of consumption and energy production on the network.

 

Approach

Electricity consumption in France is highly dependent on temperature, and this is largely due to the widespread use of electric heating in private homes and on the grid.

SYPEL (Système des Prévisions Électriques) is the reference tool for energy forecasts and meteorological data for our client. The basic need met by the project is the daily availability of energy forecasts for consumption (at the level of the 2200 source stations) and electricity production (the 7700 remote-reading producers: wind farms, photovoltaic farms, cogeneration, etc.) at the local level, and for losses at the national level.

These forecasts, which thus cover several spatial scales, are also divided into different time horizons. The SYPEL tool provides forecasts for the Short-Term (from a few hours to two weeks), for the Medium-Term (of the order of a month) and for the Long-Term (of the order of a year).

 

Key factors

  • Reconstitution of the energy consumption histories for the grid of the Source Posts from the energy extracted and injected
  • Retrieval of weather forecasts for the whole country divided into 10 weather products
  • Modelling of consumption components dependent on both temporal, calendar, meteorological and historical consumption variables using Generalized Additive Linear Models (GAM) or decision trees (XGBoost) 
  • Continuous improvement of forecast engine results using metamodels calibrated on the past error of the raw models

 

Results

  • Hourly processing and tabular integration of weather forecasts 
  • Storage of weather data since 1980
  • National consumption forecasts with an error of 1 to 2%.
  • Consumption forecasts for source stations with an average error of 6 to 7%.

 

Our teams of Data Science experts will help you in the industrialization of forecasting models. Our multi-sector expertise, particularly in the field of energy, enables us to quickly integrate the associated business constraints. In addition, our data engineers have set up a hosting platform that accelerates the industrialization of data science projects. This accelerator allows us to free ourselves from the constraints of storage or server management. 

 

Heka is the ecosystem of Artificial Intelligence solutions developed by Sia Partners. These advanced Data Science solutions come from years of development experience and support of our customers. Our developed industrial tools and insights allow Sia Partners to address recurring business issues and support value creation across multiple sectors.

 

Capability