Electrical Engineering

Site search Advanced


Using AI to improve flexibility services forecasting

Published: 12 September 2019 - Carly Wills

Smarter Grid Solutions (SGS) has used two forms of artificial intelligence (AI) to improve the accuracy of the forecasts needed to run flexibility services.

Predicting the demand and generation on the electricity network is essential to determining where and when flexibility services, such as demand side response, are required. 

SGS’s project for Western Power Distribution formed part of its Electricity Flexibility and Forecasting Systems (EFFS) project. SGS used two techniques known as ‘long short-term memory’ (LSTM) and ‘extreme gradient boosting’ (XGBoost) to make its predictions, which were more effective than ‘auto-regressive integrated moving average’ (ARIMA), the current most commonly used load forecasting tool, and techniques used in recent network innovation projects.

The results are of interest to all other Distribution Network Operators (DNOs), but particularly so for Scottish Power Energy Network’s FUSION project and Scottish and Southern Electricity Network’s TRANSITION project, which are also looking at forecasting methods. They are investigating whether the work for EFFS can be used within their own projects, reducing their costs and providing better value for money for customers. 

SGS’s project used open-source tools so that DNOs will be able to use the results widely without the limitation of needing to buy licences to proprietary software. 

SGS’s work delivered forecasts for 132kV and 33kV networks for grid supply points, bulk supply points, primary substations, large load customers and renewable generation. Timeframes from six months ahead down to one-hour ahead were examined.

Dr Graham Ault, executive director and co-founder of SGS, said: “The results from our forecasting models were very good, especially for the shorter timeframes. The accuracy levels achieved outperformed known forecasting methods used on other recent innovation projects, so these results are really pushing the industry forward. 

“XGBoost performed very well and outperformed the other methods in most test cases and applications.

“This is an exciting time for the electricity industry, with network operators making the transition from DNOs in the conventional network operator model to playing a more active management role as Distribution System Operators (DSOs).

“All DSOs will require forecasting as a core input to flexibility services procurement and dispatch, so we expect the outcomes, techniques and specific tools used in this project to be widely referred and utilised. We already have capabilities built into our DERMS software for forecasting and these results enable us to integrate ever more sophisticated algorithms, such as these AI techniques, to support our customers’ ambitions.

“The ‘how to guide’ style of project report and the open source tools will help other companies to embrace the outcomes of this work quickly and then allow them to contribute to this and other important components of flexibility services and the tools required for the wider DSO transition.”

Search for a product/supplier:
-July 2020+