Baselining
What is ‘baselining’?
In the future energy system, we’ll need consumers to reduce or shift their use to different times of the day. This will help support the grid when it is under strain. Centre for Net Zero wants to find more accurate ways to predict the energy consumption of a given household. This is known as establishing a ‘baseline’, or counterfactual.
This baseline allows us to estimate the energy consumption of a particular household on a specified day. We can use historical consumption patterns for that household in the absence of external signals to work this out. An external signal might include a request to alter their energy usage.
Why are we working on this project, and what are we trying to do?
Establishing baselines at an individual level is an active area of research. While there is a lot of existing research on forecasting individual households, there is little consensus on the best forecasting algorithm. There is also no focus on how the methods perform on households with different levels of LCT uptake, or at different times of the year. This is largely because few organisations have access to smart meter data, information about LCT ownership, or other customer metadata – unlike CNZ.
Octopus Energy has run a series of flexibility trials in the past, and our team has performed deep analysis on the results of these trials. During these trials, a baselining methodology was chosen to measure the amount of turn down achieved, and reward customers accordingly.
For the purposes of this project, we’re analysing baselines on non-trial days and recommending an algorithm we think would perform best if it were deployed as a ‘productionised’ system to model and baseline domestic flexibility. This involves looking at baseline data on non-trial days across different types of days, over several years.
Our ambition is to feed the outputs of this project into the modelling work package that CNZ is responsible for within the CrowdFlex project. This is the UK’s largest domestic flexibility study, which we’re working on with National Grid ESO and a wider consortium.
Our longer-term goal is to develop an industry-accepted, standard methodology for baselining electricity demand. This could help create an open standard for flexibility providers.
Why do we need more accurate ways to measure baselines?
Establishing the most accurate baseline is important for a variety of reasons for flexibility. Firstly, in future flexibility events, we might reward customers per-kWh deviation from their expected usage. If we ask customers to change their electricity usage, we need a way to estimate what they would have done if we had not asked them to do so. The more accurate the baseline, the better we are able to reward customers for their participation.
Secondly, we need to understand how the baseline accuracy varies at a household level. This changes according to factors such as season and low carbon technology (LCT) adoption. We can use this information to recommend the best methodology for measuring a baseline for domestic customer portfolios. This will ensure that people are rewarded fairly for any flexibility provided.
If a baseline overestimates a household’s consumption, then consumers would be given higher, harder to achieve energy reduction targets. This could potential impact opt-in rates. If a baseline underestimates a household’s consumption, then a consumer would not be appropriately rewarded for the flexibility they provided.
What are the potential wider implications of this work?
We know that the adoption of low carbon technologies is key to unlocking domestic flexibility at scale. In many cases, these technologies currently involve high upfront costs for consumers. If we can accurately measure the amount of energy saved through flexibility via low carbon technologies ownership, we can also accurately measure the carbon emissions this flexibility avoided. This can have a value in the offset market. If we can find a way to integrate that value into the financing of LCTs, then we may be able to open up new commercial models.
