The Problem

Distribution Network Operators (DNO) are facing technical and commercial challenges as the electricity industry moves rapidly towards new generation solutions and customer usage patterns. These new developments are placing stress, not only on the high voltage supply lines, but also, importantly, on the low voltage local networks as EV usage and distributed generation become more widespread. As voltage fluctuations increase, the need for accurate grid performance forecasting is becoming critical.

A DNO can use a combination of Static State Estimate and Dynamic State Estimate to underpin operational management. Static State Estimate (typically done every 5 or 15 minutes) is essentially a snapshot of the grid at any point in time. It does not register how the grid state changes over time and, because of the absence of real measured data points, it relies heavily on pseudo-measurements. Dynamic State Estimate predicts the grid state change over a relatively short time interval time (typically one time step) and is usually co-ordinated with the Static State Estimate. 

Again, the absence of real, measured data could be a problem.

 The Steelhouse solution

Steelhouse has developed a solution that brings together domain-specific, high granularity, data gathering capability and neural network/transformer time series technology to provide an innovative solution to forecasting within low voltage networks. 

Data-driven neural network and transformer time series technology solutions can capture complex temporal patterns and nonlinear relationships over long periods of time and better adapt to a continually changing environment. They are better suited to daily and intra-day real time forecasting requirements, however, the absence of current domain-specific data can be an issue, both at the training from scratch and inference/fine tuning stages.

Accordingly, for the past two years and as part of its development programme, Steelhouse has been recording real time voltage and frequency (and other power variables like power factor) on low voltage networks at multiple sensor locations in Germany and Spain. The Steelhouse data gathering program is live and ongoing and the operational database is increasing daily. 
The Steelhouse database is growing at around 250.000 data points a day and currently stands at around 150 million data points.

Using this database, Steelhouse has developed a series of neural network/transformer time series forecasting models that can provide voltage or frequency forecasts on the low voltage grid over both short and medium time horizons.

Steelhouse recognises the need to ensure that our solutions are both energy efficient and data secure. Accordingly our models have been optimised to run securely and locally on Steelhouse servers in Europe, rather than on remote, energy intensive, hyperscaler cloud data centres.