From Raw Data to Reliable Forecasts
November 4, 2025 9:04 amEvery forecast starts with raw data. At ENFOR, we build our forecasting system around two primary data streams, production data and weather data, and a third, client-specific scheduling input when available. Each of these sources plays a distinct role in shaping how the forecast is trained, updated, and delivered.
Production Data
Reliable forecasts require a strong historical foundation. We collect and use multiple forms of production data, from settlement production to SCADA readings and availability and curtailment signals, to calibrate our models.
This allows each model to learn the unique performance characteristics of every site or portfolio.
In operational setups, the system continuously updates with new incoming data. The closer to real time this data arrives, the more precisely the model reflects actual operating conditions. This is critical for intraday forecasting, where recent/latest changes in turbine availability or short-term curtailments can have a measurable impact on forecast accuracy.
Weather Data
Weather is the second main source of input. Our system integrates multiple Numerical Weather Prediction (NWP) models, each offering a different view of the atmosphere.
These models predict key parameters such as wind speed, temperature, irradiance, and atmospheric pressure.
For each NWP, ENFOR’s system builds an independent sub-model that maps weather forecasts to expected production output.
This approach allows us to exploit the strengths of different NWPs across regions and technologies, for example, one model may perform better in coastal areas, another in complex inland terrain.
Scheduling Data
The third type of input, when available, comes directly from clients. This includes operational schedules, maintenance plans, and curtailment periods that define when certain assets are expected to run at reduced capacity or be taken offline.
In ENFOR’s forecasting setup, this data can be incorporated dynamically. For example, if a wind farm operator plans maintenance on specific turbines or a solar site schedules partial curtailment during grid congestion, these events can be reflected in the forecast output. This makes the model aware of operational limitations and provides a realistic view of short-term expectations.
This type of adjustment depends heavily on the client’s use case. For many, the real value lies in knowing not just what will be produced, but what could be produced under optimal conditions. Cutting the forecast according to curtailment schedules can remove visibility of what could have been produced. ENFOR therefore often delivers forecasts showing the full production potential, allowing clients to apply their own operational limits afterward.
This approach keeps forecasts transparent and flexible, giving traders and operators a clear view of both available capacity and real production expectations.
How These Inputs Work Together
– Production data anchors the forecast to real measurements.
– Weather data projects how conditions will evolve.
– Schedules ensure alignment with planned operations
Together, they form the foundation for the forecasting process, enabling ENFOR to move from raw data to a unified, reliable forecast.
From Many Forecasts to One: The Combination Step
Each weather model tells a slightly different story about the atmosphere.
Some emphasize temperature gradients, others handle wind direction or coastal effects more precisely. Instead of choosing one and hoping it performs best, ENFOR’s system uses all of them and lets data decide how much each model should influence the final forecast.
For every NWP integrated into our platform, ENFOR builds a dedicated sub-model that maps weather inputs to expected production.
This means if five or ten weather models are used, the system produces five or ten separate production forecasts, one for each source.
Each of these forecasts represents a candidate prediction of what generation will look like in the coming hours or days. But for the client, that’s five or ten forecasts too many.
They need one clear answer!
The next step is the combination process, the mechanism that merges all individual forecasts into one final result.This is done through dynamic weighting, where each weather model’s performance history and unique contribution are analyzed continuously.
ENFOR tracks how each NWP-based forecast performs over time and how its errors correlate with others. The system then assigns higher weights to forecasts that are both more accurate and less correlated with the rest.
In simple terms:
– A forecast that consistently performs well gets a higher influence.
– A forecast that provides independent information, not just repeating others, also receives more weight.
– This creates a blended result that is both robust and adaptable, reducing dependence on any single NWP source.
Why Combination Matters
– Reduces model bias: No single NWP performs best everywhere or under every condition.
– Improves stability: If one weather model underperforms, others balance it out.
– Adapts to change: As models evolve or regions behave differently, weighting adjusts automatically, and with enough training, the combined forecast will perform better in the long term than the best of each individual NWP forecasts.
This ensemble approach allows ENFOR’s forecasting engine to capture the most reliable signal across diverse data sources, delivering one clear, data-driven forecast instead of multiple conflicting ones.
Even the most advanced weather models can’t always capture what’s happening at a specific site in the moment.
Even the most advanced weather models can’t always capture what’s happening at a specific site in the moment.
From Modeled Forecasts to Measured Reality
After the combination step, ENFOR’s system produces a unified weather-driven forecast.
While this combined forecast is highly reliable for the day-ahead horizon, it may lag slightly on the very short term because it’s based purely on weather data, it does not contain information about what the turbines or solar panels are producing right now.
A sudden change in wind direction, a passing storm front, or a local temperature variation can cause production to deviate from the expected trajectory.
That’s why ENFOR adds a real-time adjustment layer, a mechanism that continuously aligns forecasts with the most recent measured production data.
To close this gap, ENFOR introduces a short-term correction that uses live production data to fine-tune the forecast.
The correction doesn’t replace the weather-based forecast, it blends it dynamically:
- For the next few hours, the model leans more on measured data, ensuring accuracy in the very short term.
- Gradually, it transitions back to the weather-driven projection as the time horizon increases.
- This creates a smooth curve where the immediate forecast reflects what’s happening now, while the longer forecast still relies on weather predictions to look ahead.
The process is continuous, as new measurements arrive, the model rebalances itself automatically.
Real-time adjustment is critical for clients who depend on accurate short-term predictions, such as power traders and grid operators.
It ensures that forecasts not only predict future conditions but also react intelligently to the latest observations.
By anchoring every forecast to real measurements, ENFOR ensures that short-term accuracy improves without compromising long-term consistency.
Forecast accuracy is not only about how precise the models are, but also how fast new information is turned into usable forecasts.
In a market where weather changes by the minute and trading decisions depend on the latest conditions, timing can be just as valuable as precision.
That’s why ENFOR redesigned the way it ingests and processes weather model updates, making sure clients receive new insights as soon as they are available, rather than waiting for complete model runs.
