Gnarum | Empowering your energy
Gnarum’s forecasting demand module helps retailers navigate Spain’s new quarter-hourly market by combining AI/ML, ensemble learning, and high-resolution data to deliver accurate, real-time predictions for large, complex portfolios—reducing imbalances and optimizing market operations.
Starting October 1, 2025, the introduction of quarter-hourly granularity in the Spanish electricity market will increase price volatility—particularly during ramp periods associated with photovoltaic generation. This change requires retailers to significantly improve the accuracy of their demand forecasts, as any deviation from their scheduled program will now result in considerably higher costs.
This situation is further complicated by an already challenging market context: since the blackout on April 28 and ongoing grid congestion issues, the costs of balancing and constraint management services have risen significantly, increasing the price of energy procurement to meet demand.
In this environment, retailers require advanced forecasting solutions that allow them to anticipate their consumption profiles more accurately, optimize procurement strategies across different markets (day-ahead and intraday), and minimize imbalances. The ability to integrate high-resolution weather data and real-time consumption patterns becomes a critical factor for business competitiveness and sustainability.
To address these needs, Gnarum presents its new multi-resolution demand forecasting module, validated by leading retailers and designed to maximize accuracy and minimize risk in highly volatile environments.
The module combines advanced time series modeling with continuously adaptive AI/ML techniques, incorporates forecasting of key weather variables such as temperature, and applies dynamic clustering tailored to the specific structure of each portfolio. Additionally, it provides a probabilistic forecast calibrated using Conformal Prediction—essential for managing portfolios with high uncertainty.
Demand Forecasting Service for a retailer with a highly heterogeneous and geographically diversified portfolio of more than 10,000 CUPS. In particular, their portfolio changes over time and includes high-consumption supply points (6.x) with complex demand patterns.
Current forecasting models —designed for the hourly market and based on an approach focused on individual CUPS and the last recorded hourly measurement— show significant room for improvement in the new context of quarter-hour (QH) granularity. This limitation becomes even more pronounced given the structural complexity of the client’s portfolio, characterized by highly heterogeneous consumption profiles, non-stationary dynamics, and operational changes that are difficult to track with traditional approaches.
High portfolio heterogeneity
More than 10,000 CUPS with highly diverse consumption profiles.
Simultaneous operations in the Peninsula and SEIES.
Presence of all tariff types (2.0, 3.0, 6.1, and 6.2), each with its own dynamics and varying sensitivity to external factors.
High-consumption supply points (6.x) with complex profiles
Behaviors subject to seasonality.
Abrupt changes in consumption patterns driven by industrial production, scheduled shutdowns, holiday calendars, or specific events.
Differentiated Forecast Management
It is essential to forecast critical points both individually and in an aggregated manner, separately from the rest of the portfolio.
Forecast tariff groups 2.0, 3.0, and 6.X without including the critical points.
Integration of operational plant knowledge (Manual Management)
Agile manual adjustment mechanisms are required to incorporate this information without interrupting operational workflows or the traceability of forecasts.
Central busbar aggregation
It is necessary to have the full portfolio forecast ready to be sent to the Market or to each SEIES subsystem.
Any errors in aggregation and the application of loss coefficients directly translate into deviations and penalties.
Multi-day forecasting is required to monitor trend changes.
Seamless Integration with the Corporate ERP
Automatic and robust synchronization with master data (CUPS with active contracts) and actual measurements from the retailer’s ERP system is required.
The workflow must be reliable and require no manual intervention, ensuring consistency between commercial and financial operations.
System parameterization: hierarchical structure
The design is based on a two-dimensional segmentation:
Individual CUPS: supply points treated independently when specific modeling is required.
Portfolios by tariff type: groupings that allow collective handling of consumption with similar characteristics (excluding the CUPS already modeled individually).
Additionally, hierarchical aggregation levels are incorporated —Individual CUPS Portfolio and Complete Portfolio— which do not generate their own forecasts but act as summations of the forecasts from lower-level entities.
This structure is replicated across all operational markets (Peninsular and SEIES).
Dynamic portfolio management: Real-Time integration
Demand Forecast is securely and robustly integrated with the corporate management system (ERP), achieving:
Automatic synchronization of CUPS activations and deactivations.
Continuous feeding of historical consumption data.
This ensures that forecasts are always based on the current operational reality, avoiding discrepancies between the model and the actual portfolio.
To guarantee process traceability, a daily file is generated with all active CUPS.
A controlled procedure exists to properly manage the process, ensuring that when a CUPS changes status (from individual to portfolio, or vice versa), model integrity is maintained without discontinuities.
Central busbar aggregation: Regulatory and technical compliance
The system incorporates loss coefficients published by REE for each tariff type and market, in order to convert the forecasted demand at the supply node to the demand at the central busbar—an obligatory requirement for participation in wholesale markets.
Service modeling and operation
The algorithm dynamically calibrates characteristics and learning techniques to optimize the forecast for each entity, incorporating not only the usual variables for electricity demand forecasting but also a wide range of meteorological variables, national, regional, and local calendars, marketing calendars, among others.
In particular, each individual CUPS is assigned a specific learning algorithm with its own time series. The system detects pattern changes and adjusts the model through continuous and adaptive training, leveraging its hourly-updating Telemetry system.
Manual Forecast Integration
For special cases, the system allows the automatic forecast to be overridden with a manual input (e.g., an industrial client with an exceptional scheduled shutdown).
The manual forecast takes priority for the specified time period.
Both versions (automatic and manual) are recorded and displayed in reports, ensuring full traceability of the process.
Thanks to this flexible parameterization approach, seamless integration, and the accuracy of its forecasting algorithms, Forecast enables retailers to significantly reduce deviations and optimize their position in the day-ahead and intraday markets over a multi-day horizon.
