By using a combination of special purpose, network sensors and neural network time series technology, Steelhouse Data Networks is able to provide time critical voltage and frequency forecasts. 
The models are based on, recently developed, 
neural network and AI transformer technology. Various product sets are available and Steelhouse has developed and adapted forecasting models that can use zero shot forecasting (full zero shot inference or with parameter efficient fine tuning) and models that can be trained from scratch. Training from scratch typically involves millions of domain specific data points. The models can be fine tuned or retrained if accuracy drift is detected.

VOLTAGE FORECASTING 

The Steelhouse technology can be applied across multiple low voltage electricity grids allowing more efficient operations for a DNO.

The context period and horizon can be adjusted depending on DNO requirements and can range from minutes, for „within control cycle“ operator input, to days, for planning purposes.

Only a small amount of anonymized data from the DNO 
is required as context for an accurate forecast.

GRID FREQUENCY FORECASTING

Steelhouse has also built and trained a series of grid frequency forecasting models. 

Since October 2025 energy (kWh) is traded on a 15 minute basis, with large amounts of energy being brought onto the grid and taken off the grid every 15 minutes these are the boundaries or cusps when frequency
control is at maximum stress.

We offer a dynamic frequency forecast for the next 3 to 6 cusp values. 
 

Please contact us if you wish to discuss any of our products.