Energy forecasting in the new normal
Updated: Feb 4
As the energy industry emerges from COVID-19 lock-down it needs to mitigate balancing and bad debt risk. Good forecasts of demand and payment are essential. The answer is auto-renew models.
Energy & Lockdown
According to Gridwatch, (www.gridwatch.co.uk) Gas demand has plummeted from 125 GW/day pre lockdown to 75 GW/day post lockdown.
The chart shows total gas flows as reported by National Grid on the National Transmission system. The red slither at the top of the chart is large directly connected industrial load, the orange is gas fed to the local gas distribution networks, and the blue is gas fed to power stations.
Electricity demand has also slumped from 35 GW/day to 25 GW/day.
The top slices down to the pink are the interconnectors bringing in electricity to the GB market. The yellow band is solar generation. The turquoise underneath it is wind power. The thin black line underneath that is coal fired generation – now virtually zero in GB. The thin dark brown line is Biomass generation, the large orange band is gas fired power station supply, and the grey at the bottom is nuclear generation.
And this hides even greater changes at the sector level as commercial industrial demand has borne the brunt of the downturn as factories, shops, bars and restaurants have closed. Heating demand for homes has dropped with the warmer weather, but given home working, furloughed staff, and the stay at home regime - home occupancy is higher which will have fed into higher gas and electricity demand in the residential sector.
The big question is what will happen as we move cautiously out of lockdown?
Rising Imbalance & Payment risk:
Suppliers and traders face two big lockdown emergence risks.
The first is imbalance charges as they fail to estimate the offtake that their customers will require as they switch to life in the new normal.
The second is counter-party risk. For a supplier this is the heightened risk that their customers will be unable to pay their bills. In the business customer world this will include businesses that do not re-open, but also those that do but then find that they are not viable in the new world. In the residential world there will be households who lose their income. For traders this will be counterparties who default on contracts.
Current modelling tools are inadequate for handling this new and rapidly changing world because they are either calibrated on the historic data of the old world or built on assumptions, which are probably wrong, about how the new normal operates.
Indeed, this week’s model based on data up to and including yesterday will not be suited for next week’s forecasting task since the world will be significantly different.
So how can energy suppliers and traders assess and manage the lockdown emergence risks that will arise as businesses come to terms with the impacts of Covid-19.
Lockdown emergence risks can be mitigated by deploying auto-renew models operating on new normal data to provide actionable insight
This is because:
1. Auto-renew models draw insight from the latest data which contains within it the knowledge of how the world has changed and is changing. This is achieved because Auto-renew models rebuild themselves from scratch as new data becomes available. This could be hourly, daily, or weekly or any appropriate time period.
2. Auto-renew models provide insights on how the world is changing This is because they can report on the balance of predictors used – which variables, and for which time periods. This is a source of insight and knowledge of how relationships function in the business system, and how they are changing over time. This knowledge of how the business system works can then be used for modelling potential future worlds as would be done in scenario planning.
3. Auto-renew models are available real-time since they can build and optimise themselves in seconds In an auto-renew model environment models are built and optimised in a single pass through the data that they are interrogating. This delivers incredible speed in the model building and optimisation process. It is possible to build, optimise, and run a model to create a forecast within between 30 seconds to 2 minutes using RTinstantML (real time instant machine learning). An artificial-intelligence based form of computation.
4. The incremental cost of building and running auto-renew models is virtually zero. The marginal cost per new model is measured in pence not £’s, once the auto-renew model building framework is in place. Once you have the system in place whether you build 10, 100, or 1000 models the cost does not materially change. This is very different from traditional model building approaches where teams of data scientists and programmers are needed to build and optimise each model. However, this does not mean that data science teams are not needed. They are – but for the value adding aspects of model creation. Human intervention is needed for interpretation and decision making based on the model outputs. However, the model-build and run is fully machine driven. In addition, human data scientists are needed to set up the automated auto-renew modelling systems and manage the data feeds to the system.
If you would like to find out more about auto-renew models please get in touch or take a look at www.tangent.works an automatic model generation company with whom we have been working with for over a year.
We would be pleased to explore with you how this capability could save you significant sums of money in managing imbalance risk and bad debt as we emerge from lockdown.
All a conversation costs is a few minutes of your time – but it may save your company millions of pounds in the transition to the new normal.