Eneco increases energy forecast accuracy by 10% using machine learning

The energy landscape has undergone a radical transformation due to the rapid adoption of solar panels, electric vehicles, and heat pumps, combined with price volatility driven by global events. Traditional planning tools proved inadequate for a sector where energy consumption had been predictable for decades. As an energy supplier, Eneco must accurately forecast customer energy consumption and procure energy from grid operators to ensure grid balance. Forecasting errors entail costly consequences and pose risks to grid stability, making the improvement of long-term planning essential.

Eneco launched a large-scale project with the support of BigData Republic to develop automated systems and machine learning-based forecasting models for gas and electricity. The initiative was rolled out in phases:

• Establishing a benchmark: replicating existing forecasting methods for gas and electricity as a baseline for comparison)

• Proof of Concept (PoC): developing a forecasting model for a pilot group of 200 industrial connections with at least three years of historical data

• Proof of Value (PoV): scaling forecasts to all industrial consumers via a dedicated pipeline featuring optimized data flows and parallel processing)

• Automation and integration: integrating forecasts into the operational platform for daily use by sourcing and pricing teams, supported by continuous long-term monitoring.

To build trust and acceptance among end-users, the teams were closely involved in the development process through the use of interactive charts, examples, and open communication regarding progress.