How AI is shaping the future of energy

Michael Bironneau, Technical Director of Open Energi explains how artificial intelligence can be used to provide demand side flexibility more efficiently.

 Technology is changing the way we live. By 2020 it is estimated that there will be over 50 billion smart devices connected to the internet, creating an ever-expanding world of communication and data flows. Rapid advances in artificial intelligence (AI) and machine learning (ML) mean this digitised society will be able to run itself more smartly and more efficiently than we could ever have imagined.

As we shift to a decentralised, decarbonised energy system, this intelligent connectivity can be leveraged to orchestrate massive amounts of demand-side flexibility, enabling supply and demand to interact intelligently to deliver a flexible, zero-marginal-cost energy system.

At Open Energi, we’ve been exploring how the same mathematical techniques that have let machines defeat chess and Go masters can help to radically reduce the cost of consuming and delivering power.

Unlocking value from multiple assets

Over the last 7 years we have worked with some of the UK’s leading companies to manage their flexible demand in real-time. In the process, we have connected over 3,500 assets at over 350 sites, operating invisibly deep with business processes, to switch equipment on and off in response to fluctuations in electricity supply and demand.

Already, we are well on the way to realising a smarter grid, but to help businesses unlock the total value of their demand flexibility – across all their assets – we need to adopt a portfolio level approach. The challenge lies in finding ways to automate the behaviour of many disparate – and potentially business critical – assets, from industrial equipment through to battery storage systems, without disrupting business performance. Because each asset is likely to have a primary purpose other than providing flexibility, there are bespoke and dynamic operational constraints that must be considered.

Relying on a traditional mathematical model – based on a linear system – to manage this portfolio of assets is not feasible. The manpower required to encode these would be phenomenal, not to mention the computing power that would be required to solve it. A more real-time approach is required.

Using AI, we can create an efficient and adaptive framework that can look across multiple assets on a customer’s sites and, given all the operational parameters in place, intelligently optimises their behaviour so that they consume energy in the smartest way to save energy, cut costs and earn revenue.

Take a supermarket which has solar panels on its roof and a battery installed on site, as well as flexibility inherent in its air-con and refrigeration systems. Operating these assets in isolation and optimising one store, rather than looking across a portfolio of stores, would miss huge opportunities for savings and income.

The goal is to maximise value for a business subject to local and portfolio level constraints, which could mean fulfilling a Capacity Market contract with flexible demand from a portfolio of stores while minimising the impact to consumers and continuing to track grid frequency on a second by second basis for frequency response.

An AI model can make decisions in real-time about how all these sites and assets should be managed to deliver the most value; when, where and how much flexibility exists, when to charge and discharge the battery, and when to use, store or export energy generated by the solar panels.

Similarly, imagine a customer who has done a portfolio-level power purchase agreement (PPA) with its supplier or is otherwise exposed to a portfolio-level balancing cost. At a given time, one site may have more flexibility than another, and therefore it may be more financially rewarding for that site to use its flexibility to reduce the imbalance of the customer with regards to the PPA, effectively “shifting” it to the other site, rather than do something else, e.g. dynamic frequency response.

AI means we can find creative ways to reschedule the power consumption of many assets in synchrony, helping National Grid to balance the system while minimising the cost of consuming that power for energy users.

 Powering an AI model

Lack of data is often an obstacle to progress but we collect between 10,000 and 25,000 messages per second relating to 30 different data points and perform tens of millions of switches per year, operating invisibly deep within business processes.

This data is being used to train a deep learning model which combines asset-level constraints from a bottom-up approach with portfolio-level modelling that can tweak the outputs to improve the aggregated solution.

AI model learning to control the electricity consumption of a portfolio of assets


The model can look at a sequence of actions leading to the rescheduling of power consumption and make grid-scale predictions saying “this is what it would cost to take these actions”. The bleeding edge in deep reinforcement learning shows how, even with very large scale problems like this one, there are optimisation techniques we can use to minimise this cost beyond what traditional models would offer.

More rapid progress could be made across the industry if energy companies made more anonymised half-hourly power data available. It would enable companies working on smart grid technologies to validate these ideas quickly and cheaply. In the same vein, it would be a major breakthrough for balancing electricity supply and demand if energy companies made available APIs for reporting and accessing flexibility; it would allow companies like Open Energi to unlock enormous amounts of demand-side flexibility and put it to good use balancing not just the grid but also helping to optimise the market positions of those same energy companies.

In the UK alone, we estimate there is 6 gigawatts of demand-side flexibility which can be shifted during the evening peak without affecting end users. This is equivalent to roughly 10% of peak winter demand and larger than the expected output of the planned Hinkley Point C.  AI can help us to unlock this demand-side flexibility and build an electricity system fit for the future; one which cuts consumer bills, integrates renewable energy efficiently, and secures our energy supplies for generations to come.