Building virtual storage in electricity
_{is there a shortcut?}

Needed: Intelligent Local Grids. When: Immediately

The push to NetZero places most of its bets on electrification[1]. While, on average, total UK electricity demand has been decreasing since 2005[2], massive changes are brewing. Increases in distributed generation and demand due to the electrification of heat and transport are fast approaching.

These changes will require much more ‘intelligence’ from low voltage local networks to deliver more flexible load management if they are to continue to function within operational and regulated limits, and maintain grid stability. Such flexibility management at the local level must also be designed to benefit from the wholesale electricity and balancing market, otherwise the investment in flexibility will be under-optimized.

Increased visibility through monitoring will be the key to unlocking new intelligent solutions and keeping costs down. One of the benefits of more data is to facilitate more reliable forecasts of both demand and supply, which in theory can enable the operators to act more optimally. Smooth demand curves have a double advantage: they postpone the need for expensive reinforcements or negotiating massive reserves to be able to meet peak demand, and they allow for better prediction.

As we know, demand smoothing is difficult, but it can be leveraged through time-of-use tariffs, by automating demand shift (e.g., through smart charging or intermittently switching off appliances), or by using storage[3].

So, the solution is mostly…storage?

Storage requires the least amount of behavioural change, but as we know, it is expensive. An alternative approach would be to create virtual storage, through flexibility. By ‘flexibility’, we assume a timely and/or spatial shift of activities, that reduces demand during peak periods, moves it from substations with a little headroom to those with more headroom or from less green to a greener generation mix.

But, to shift demand, we need to be able to predict it accurately.

New sources of data can improve the prediction of local demand: mobility, transport, local weather forecasts, smart meters, etc. Still, we are dealing with complex human interactions, where the only constant is change. For example, the future equivalents of the TV pickup[4] for soap operas would be difficult to predict – as content becomes available 24/7, and one can pause broadcasting, synchronous breaks are rarer.

A complex system of a dynamic environment and human behaviour presents us with various challenges:

• climate change impacts time-of-year usage patterns and the frequency and intensity of peak electricity demand[5];
• shifting working patterns caused by COVID moved demand from industrial buildings to individual households and disrupted daily time-of-use patterns;
• different policy considerations risk the creation of new peaks (e.g. an increased early evening peak, or localised night peaks for EV charging);
• technological innovations (cryptocurrency, large language models) generate new significant electricity demand. 

Follow the yellow brick road…

Predicting accurately and then shifting demand equates to building robust virtual storage solutions, but obviously, the devil is in the detail.

We can think about three different levels of time/space demand shifting:

• Macro-level – Large consumers (e.g., data centres[6] and other industrial consumers)
• Meso-level – Aggregated flexibility services
• Micro-level – Individual households, secondary substations, or distribution feeders[7]

The different levels present different technical challenges when trying to create data profiles of these entities and ecosystems.

Individual large consumers or categories of commercial consumers are, in general, easier to predict as they will have smoother profiles[8]. But because they also have harder constraints, need more time to react, and have a vast range of data, the challenge is higher. That is why it is much easier to model domestic demand over commercial or industrial.

On the other hand, flexibility at the meso and micro levels allows for more adaptable solutions. However, while they might be able to act faster, accurate forecasting is much more difficult at the feeder or individual level due to the volatility and variety of behaviours, not to mention privacy and scalability issues. What’s more, local dispatch optimisation problems can become too big to solve simultaneously.

The answer, therefore, may lie in the meso-level. Creating data-profiles of similar neighbourhoods and similar businesses[9], and aggregating these, could provide an innovative ‘network of small virtual plants’, providing a storage solution that is robust, flexible, and cost-effective. By creating virtual storage, the UK can manage predicted increases in electricity demand, and thus improve its odds of meeting its Net Zero pledge.

[1] https://www.iea.org/reports/net-zero-by-2050

[2] https://www.gov.uk/government/statistical-data-sets/historical-electricity-data

[3] N.B. that the technology is still some distance from scaling to the level needed

[4] https://en.wikipedia.org/wiki/TV_pickup

[5] https://www.pnas.org/content/114/8/1886

[6] Our data centers now work harder when the sun shines and wind blows (blog.google)

[7] https://doi.org/10.1016/j.erss.2019.02.008

[8] https://arxiv.org/pdf/2106.11750.pdf

[9] https://www.creds.ac.uk/publications/development-of-a-profile-based-electricity-demand-response-estimation-method-an-application-based-on-uk-hotel-chillers/