Net zero electricity systems will be based on AI-Optimised Electricity Markets
In their proposal for the UK’s 6th Carbon Budget, the Climate Change Committee (CCC) recently recommended that the UK seek to reduce emissions by 78% by 2035. They estimated that to achieve this the UK will need to have almost completely decarbonised the UK electricity system and at the same time annual electricity demand is going to increase from 300 TWh in 2020 to 460 TWh by 2035. They posed a scenario where 70% of generation is renewable by 2035, with the bulk of this (265 TWh) coming from offshore wind. It is clear that large amounts of flexible balancing capacity will be needed to manage such system requirements and large public and private investments are being made in the physical assets that can offer balancing capacity.
The key question now is how can all of this balancing capacity be managed and delivered without compromising grid function and with minimal cost for the consumer? The answer is that in addition to all of the physical assets required, we will need AI-optimised energy markets to allow the new energy system to operate efficiently and deliver affordable energy for consumers.
The development of the wholesale electricity market and balancing markets has been shaped by the dominant sources of energy generation, which are now rapidly changing. The design of electricity markets will need to adapt quickly to manage new generation and flexibility sources.
Whilst it is increasingly clear that the UK’s future generation profile will involve large amounts of variable renewables it is less clear what the flexibility mix will look like. The potential outcomes vary widely. On the one hand we could imagine a system where we have vast numbers of households and EVs offering balancing services, requiring accurate real-time price signals. On the other hand it may prove to be the case that the most cost-efficient way of delivering balancing services involves fewer, larger storage and DSR assets strategically positioned at key locations on the networks, together with much simpler price signals to consumers to, for example, charge EVs at night.
Provided the rules are set in the right way, ultimately the market will determine which approach is the most economic. But regardless of the outcome of this issue, the key question that needs to be answered is how can we ensure the optimal use of whatever balancing mix wins? The answer is that optimal balancing of supply and demand will require AI optimised markets. What do we mean by this?
We mean that all buyers and sellers on the electricity markets that are able to supply balancing services will need to use AI optimised trading algorithms, AI-optimised forecasting algorithms, and in some instances AI-optimised asset planning. Generators will need to optimise their selling of electricity against the market. There are already examples where AI has been used to increase the value of wind assets. This will become the norm. Suppliers will need to optimise their buying of electricity from generators and flexibility providers on the wholesale market. Once again, this is already happening to some extent with Octopus and Ovo leading the way in the UK. Flexibility providers and aggregators will need to use AI to optimise their trading against their assets’ internal chemistry / physics, and against the markets into which they are trading. For example Habitat Energy is using AI optimised trading algorithms to optimise its battery trading against grid batteries’ chemistry and against the wholesale and balancing markets. Networks (ESO and DNOs) will need to optimise their purchasing of services with real-time auctions for balancing services, by allowing for optimal algorithmic trading into these markets by flexibility providers and by supporting data flows to ensure optimal, cost-efficient balancing.
AI-Optimised trading algorithms will in turn need good data: good data on any asset(s) that they are offering services against, and good data on the wider market. They will also need to access optimised forecasts of both demand and supply, which in turn will need to be optimised by AI. Such forecasts will become increasingly important as percentages of variable generation on the grid increase.
What we are seeing is that market participants at the grid edge are increasingly deploying AI to optimise their trading performance. However, they can only work with the data and trading environment that is currently available. To successfully offer the balancing services needed to deal with much larger proportions of variable renewables, they will need access to data that can radically improve their understanding of current and future supply and demand across the system, and a more technologically up to date process for trading on the electricity markets.
There is an increasingly important role for the network companies to facilitate the development of this system. This is a critical piece of the puzzle if we are to achieve efficient net-zero electricity grids.
So how do we get from where we are now to AI-optimised electricity markets? Here are some suggestions on the immediate priorities:
All market participants will need to continue to optimise their trading strategies using machine learning, and clarify the data streams that will allow them to do this.
Network companies could usefully focus on facilitating optimised purchasing of balancing services. To achieve this they will each individually need a far better understanding of all of their assets’ location, requiring a digital map of assets, real time data collection from across their networks, and much better forecasting capabilities of supply and demand. Network companies should consider developing roadmaps towards achieving AI-optimised balancing markets and consider working together to improve their demand and supply forecasting capabilities.
Policy makers in government need to ensure that energy data that will allow for improved understanding of current and forecast demand, supply and constraints across the system is opened up to allow for more efficient balancing and optimised planning for future assets.
Electricity markets need to move closer to real-time trading, exchanges need to offer APIs to improve market optimisation and agree on common data frameworks to make it easy for assets to easily trade on multiple markets. Simpler markets are easier to optimise.
Network companies need to rethink how they develop software, quickly moving towards a more agile and iterative way of developing software.
We will need all of these changes in place by 2035, but why wait? The UK is already a leader both in the transition of its energy system to net zero and in AI. We could be a world leader in AI-optimised energy markets. Wouldn’t that be a good story to tell at COP26?