Flexitricity Scales Energy Forecasting MLOps to Drive Value in Hours
About Flexitricity
Founded in 2004, Flexitricity is Great Britain’s first pioneering demand response aggregator, which has grown to support 1.1GW+ of fully flexible virtual power plant (VPP) assets across the country. Flexitricity is helping to build a greener energy network, by enabling industrial and commercial businesses, investors and developers, utility companies and public sector organisations to earn revenue with flexible assets.
1st Demand Response Aggregator in GB
1.1GW+ Contracted VPP of Fully Flexible Assets
24/7 Control Room and Operations
The Challenge
Flexitricity is a highly technical demand response business managing an extensive VPP flexible asset portfolio and energy trading team. It is developing an updated offering for its energy partners enabling greater value capture for behind-the-meter assets, aligning with changing market access regulations within the GB energy market. Flexitricity needed to further scale its machine learning operations (MLOps) without growing the internal data team unnecessarily. The requirement was time-sensitive, as scalable MLOps needed to be implemented and robustly tested to ensure viable development of Flexitricity’s evolving energy offering. The team required MLOps expertise to complement and support its internal data capability, to mature and establish best practice energy forecasting MLOps.
"Getting a model into production can now be done in an hour. Having a consistent process really drives that speed improvement.”
Steve Sinclair
Head of Data at Flexitricity
Our Approach
Hypercube engaged with Steve Sinclair, Flexitricity’s Head of Data to design a machine learning modelling pipeline that would enable his team to produce energy usage predictions faster and at scale. The complex models need to ingest data from a variety of sources, including IoT devices on a range of flexible energy assets, and need to predict energy usage every 30 minutes.
The Flexitricity and Hypercube teams collaborated with Miro to map out the architecture required and Hypercube began building Azure MLOps pipelines within Azure. Dummy data was used to iteratively improve the models until forecasting and prediction performance was optimised for use in real-world environments. Hypercube built Azure machine learning capability with best practice approaches, automating the process of ML creation, from raw data into reproducible ML models. Hypercube also laid the foundations that enables model monitoring for drift detection.
The automation transformed the time it took the Flexitricity team to productionise energy forecasting ML models; the process now only takes hours. This optimisation has been transformative for the team, and the time saved has enabled Flexitricity to increase the experimentations it can run to thousands each year – unlocking huge potential for advancements in the data science team and wider business. The scalability of the standardised models and re-useable code has enabled Flexitricity to further develop its behind-the-meter energy offering with confidence.
Outcomes
ML Models Built in Hours
Flexitricity now has best-practice, automated MLOps pipelines that have reduced production time.
Achieves Executive Objective
Scaling the MLOps practice was a business imperative for Flexitricity's offerings, working with Hypercube enabled the team to meet the needs of the executive board.
Increasing Data Science Experiments to 100000+ Per Year
ML optimisation has transformed the potential of experimentation for the internal team.
Secure by Design
Built on Azure ML, the models are secure by design, with governance guardrails and user roles access implemented and defined.
Hypercube supports energy asset owner-operators, traders, route-to-market providers, and energy services companies to achieve world-class AI, data and analytics capabilities.
"The number of experiments we can run has jumped to hundreds of thousands a year. From a data science point of view, that’s really important. From the business point of view, there’s so much more confidence that we can scale with the VPP and that models will make it into production quickly. You can start driving value far quicker, which is the key thing for me.”
Steve Sinclair
Head of Data at Flexitricity
Download the full case study
Join the Beyond Energy community
Invitations to exclusive events, energy leader roundtables, webinars and on-demand videos, and the Beyond Energy newsletter.
