Improve the ability of GE solutions to predict electricity consumption and production at local scales and short term, taking into account the intermittency of renewable sources and consumer behaviours.

Use case

Forecasting at short timescales is a critical process in power system operations, in particular, when it comes to integrating an increasing share of intermittent renewable energy resources. In this process, utilities are assisted by forecasting software which presents them with forecasts of the energy production and consumption on the grid, thus offering them the possibility to look ahead and optimize their grid operations.

The quality of forecasting goes along with how efficient the utility is able to dispatch power across the grid at an optimal cost and/or minimal carbon emissions, while accommodating the grid load capacity.

The statistical forecasting techniques work well at the national level, but their relevance can drop down to only 30% at local perimeters.

Traditionally, forecasting techniques rely on statistical analysis to compute the forecast of the energy production and consumption. The combination of load and wind or solar forecasting enables the utility to create the net load uncertainty that must be managed by the utility economic dispatch process or with suitable reserves. This technique works well statistically providing forecasts at national or regional levels for a mass of assets, with confidence levels up to 95%. However, it shows large deviations when it comes to computing the forecast at a small local perimeter, say for a solar district or for an individual consumer for example; confidence levels may be as low as 30%.

Challenge objectives

This challenge is about improving the forecasting capability at local perimeters, of two kinds of information:

  • The power production from wind and solar, which is impacted by weather factors.
  • The power consumption by customers, whether individuals or businesses. It is impacted by individual behaviors, and this impact could be better anticipated if direct information such as that from smart meters associated with indirect information such as that from social networks, connected objects, or other sources, was used as input in prediction algorithms.
Forecasting can be improved by taking in input new information such as weather factors or data from social networks or connected consumer devices.

Typically, such new forecasting capability will enable the utilities to perform more precise power dispatch (control of generation side, or storage, or flexible loads) surgically, for example targeting a congested zone, or to optimize their power dispatches continuously in order to accommodate near-real-time expected variations.

Why Predix?

The new forecasting algorithm is a good candidate for integration within the Grid IQ Insight product of GE Grid Solutions, powered by Predix. This software aims at monitoring and analyzing the outputs of Distributed Energy Resources (DER) connected to the grid (i.e. distributed generation, storage and flexible loads). Below is a screenshot of the look and feel of such a tool.

Grid IQ Insight: a candidate where the new prediction algorithm may be integrated.

Grid IQ Insight: a candidate where the new prediction algorithm may be integrated.


Energy injection and consumption history

The challenge partners team with GE Grid Solutions has identified one of their clients as interested in experimenting the forecasting solution. They own the historical data on which the algorithm could be applied:

  • Injection curves: historical time series of wind and solar energy injected into the grid, as produced by on-site smart meters;
  • Consumption curves for each individual customer of the experimentation perimeter.

This client will be onboarded during the prototyping phase and share the data at that moment.

Extra data

Any additional data meaningful to the energy curve prediction can be brought by the candidate team, depending on their expertise field: weather for the injection forecast, social networks or IoT for the consumption forecast, etc.

Success metrics

Desired outcome if the problem is solved

The more local the scale at which the grid management is handled, the easier it is for utilities to deploy new energy services and innovation at the grid edge.

Today, at national level, the forecasting techniques may reach a confidence level in the order of 95%, while at local level it may only reach 30%. The objective here is to make the small-scale prediction as good as the larger-scale forecasting performance.

How will it be measured?

The performance of a predictive algorithm is measured as the time-averaged discrepancy between the predicted and the realized signal curve. Success here will be assessed by computing the confidence level resulting from the use of the new algorithm.


This challenge is sponsored by GE Grid Solutions.

Said Kayal Smart Grid Innovation Director Coordination

Said Kayal

Smart Grid Innovation Director


Frédéric Héliodore R&D Project Leader Data Science Expertise

Frédéric Héliodore

R&D Project Leader

Data Science Expertise

Rodolphe de Beaufort Marketing Director Client-side Business

Rodolphe de Beaufort

Marketing Director

Client-side Business

Philippe Cros Partnership Director Startup-side Business

Philippe Cros

Partnership Director

Startup-side Business

Business collaboration scenario

The preferred approach is a licensing agreement, whereby the team offers GE an x% rebate on the software or microservice, which can be integrated to GE Grid’s software environment presented to clients.