Fault Forecasting: Improve Fault Event Detection for Distribution Network Operators

Distribution Network Operators (DNOs) face daily planning and operational challenges due to the weather. Reliable weather forecasts are, therefore, vital for them to ensure the power supply anywhere, anytime.

 

The fault forecasting model leverages sophisticated machine learning technologies to translate weather forecast data into a numerical prediction of power network fault volumes that vary with time and geographical location. The model has been developed in close collaboration with UK-based DNOs and helps them to deal effectively with power outages.

 

Why was the fault forecasting model created?

 

Traditionally, DNOs prepare for severe weather situations by consulting weather information, which they then manually translate into an estimation of fault volumes. Although the operator's expertise is essential, it is not infallible due to the complex relationship that the weather has with fault volumes. The fault forecasting model, however, may detect fault events that may go unnoticed by looking at weather data alone. Therefore, it helps DNOs with more efficient planning for resource deployment, reducing mobilization costs, and improving restoration times.

 

How was the fault forecasting model created?

 

The fault forecasting model is based on historically observed fault patterns and, therefore, able to accurately translate weather data into expected fault volumes. The fault forecasting product consists of a cloud-based modeling engine and a web-based front end for customer interaction.

 

The fault forecasting model engine 

 

The model engine uses a tree-based machine learning technique that combines historical fault data provided by the DNO with MeteoGroup's proprietary weather forecast data (MOS). Depending on the client, the machine learning may incorporate more than 10 years' worth of data.

 

The MeteoGroup MOS weather forecast feeds the engine to predict the expected number of faults. Fault predictions currently have a temporal resolution of 6 hours and cover a 5-day forecast horizon. However, the model can be configured to allow for higher temporal resolutions or any other license (sub-)area, as requested by the DNO.

 

What is MeteoGroup's MOS System?

 

Model Output Statistics (MOS) is a real showpiece at the heart of creating your forecasts. MOS adds value to raw weather model output by correcting for local conditions that are present at your point of interest. The statistically corrected MOS forecast is, therefore, more accurate than the raw weather model.

 

At the core of the MeteoGroup MOS lies the mathematical technique called multiple linear regression. The MOS contains hundreds of millions of equations that are trained by comparing historical local weather observations with raw weather model data. The MOS realigns every systematic deviation and applies the appropriate weighting and attributes from multiple weather models. It thereby corrects for local conditions and phenomena that are not represented well in coarse-scale weather models. As such, every weather parameter for every observation point generates a specific individual MOS forecast.

 

Fault forecasting web-based front end

 

The fault forecasting dashboard visualizes the predictions and allows the client to prepare and plan their response to severe weather situations. The dashboard gives a geographical depiction of the DNO's license area overlaid with the current fault predictions. While this interactive map allows for zooming and panning, the fault predictions can also be visualized in tabular form. Additionally, various weather parameters are available as interactive layers that can be toggled on or off, for example, wind speed or direction.

 

The forecast updates on an hourly basis, with a lead time up to the next 5 to 10 days. MeteoGroup experts ensure forecast quality through verification studies and manual backtesting.

 

To find out more on how we can help you please get in touch with us.

 

Contact Us