Arrow DownArrow ForwardChevron DownDownload facebookGroup 2 Copy 4Created with Sketch. linkedinCombined-Shape Combined-Shape mailGroup 4Created with Sketch. ShapeCreated with Sketch. twitteryoutube

Case study

Regulating the pressure in the UK gas grid and reducing carbon emissions

We built a model to accurately predict pressure demands in gas. Using our model, we were able to reduce the standard error by 52% and significantly reduce uncertainty. Our model could save the supplier £2.5 million per year.


An innovative startup aiming to transform the UK gas grid.


Leaks of methane gas are common in the UK gas infrastructure. Because methane gas itself is a greenhouse gas 80 times more potent than CO2, those leaks (which increase as the pressure within the pipes increases) also have an environmental cost.

There are competing forces at work: the UK Government is encouraging suppliers to reduce pressure in their gas supply (to meet climate-change goals), but if the gas pressure is too low, there is a greater risk of explosions (for example, if pilot lights on ovens fail). Our client has created hardware that allows the pressure within gas networks to be controlled remotely. We were asked to develop a model that could accurately predict demand for gas.


We reviewed data from a local gas network, and established the pressure difference between the main inlet and the low point of the network. The aim was to reduce this pressure difference as much as possible.

Left unregulated, the difference has two peaks each day: in the morning and in the evening.

Graphs of the error over time of Faculty’s model (red) and the industry standard model (blue). Faculty’s model made predictions with significantly less error.

The industry standard uses a rolling mean that captures these peaks, but there is still a significant discrepancy between pressure predicted using the industry standard model, and the actual demand. The discrepancy is caused by a failure to take into account periods when temperatures are either lower or higher than average. On warm days the usage of gas is overpredicted, whereas on cold days it is underpredicted.


Using our model, we were able to link gas usage to temperature and significantly reduce uncertainty. We reduced the standard error by 52% compared to the industry standard.

It is possible to put a price on the potential benefits of using machine learning to predict pressure demands, calculated by reference to government incentives to suppliers. Our model could save the supplier £2.5 million per year.

Applied to the gas network across the UK, our model could save £10 million and prevent the release of 15,000 tonnes of methane gas, a reduction equivalent to taking 600,000 cars off the road.


in annual savings


reduction in
standard error

To find out more about what Faculty can do
for you and your organisation, get in touch.