Network

NETWORK / 28 / FEBRUARY 2019 GAS their own forecasting tools in place, fore- casts can be limited by current technology. NGN and Newcastle University have demonstrated how these new forecasting methods work in practice. Why the need for accurate forecasts? NGN supply gas locally from their o akes connected to the national network into homes and businesses throughout the north and north eastern regions of the UK, and are required (at the most detailed level) to forecast gas demand and supply on an hourly, live basis. Balancing supply with demand is chal- lenging for the industry, especially if there is a lack of available data to make accurate demand forecasts, or if the data itself is unreliable e.g. poor weather forecasts. The network deals with unexpected demands causing spikes in the network, such as a sudden change in weather, resulting in too little or too much gas in the network. There is also the impact of new connec- tions such as the recent rise in the number of electrical "peaking plant" generators creating demand pro€ les di‚ erent from more traditional industrial connections. The short burst operation of these plants has an impact on the local storage position. Gas distributers oƒ en have to redirect gas on an hourly basis to cover unexpected demands. This impacts both system e„ ciency and … exibility, and can lead to cost implications if not corrected using standard gas balanc- ing tools. Any forecast error is compensated for through the application of storage and modi€ cations to the supply position. An over forecast will lead to the storage system increasing, or if the forecast is too low, decreasing. Such errors across the wider UK system lead to impacts on the national stor- age position and balancing actions being taken to maintain daily strategy. All of this results in less e„ cient systems, higher costs and reduced … exibility overall. A good forecasting process requires more than taking data samples and using them to project future demands based on the past. Dr Kevin Wilson, Lecturer in Statistics with John Quigley, Daniel Henderson, Malcolm Farrow, Sarah Heaps & Matt Linsley (School of Mathematics, Statistics & Physics, Newcastle University) Bayesian Statistics Application in Industry Demand Level Demand Level Henderson, Malcolm Farrow, S Heaps & Matt Linsley (Scho Sarah Actual demand Forecast median Demand Level Demand Level 01 Jan 06:00 12:00 18:00 02 Jan Time Demand Level Actual demand Actual demand Actual demand 01 Jan 06:00 12:00 Time Demand Level 12:00 18:00 02 02 Jan Time 01 Jan 06:00 12:00 18:00 02 Jan Time Forecast median orecast median 01 Jan 01 Jan 06:00 12:00 Time 18:00 02 Jan Site: X | Last Updated: 2015-01-01 00:00:00 Site: X | Last Updated: 2015-01-01 04:00:00 Site: X | Last Updated: 2015-01-01 08:00:00 Site: X | Last Updated: 2015-01-01 012:00:00 Site: X | Last Updated: 2015−01−01 16:00:00 Site: X | Last Updated: 2015-01-01 018:00:00 Strategies must also be included which help predict how much gas should be supplied at a particular time using historical and live data, alongside process experience and expertise – a key bene€ t of Bayesian-based forecasting. Underestimating or overesti- mating demand will result in an incorrect amount of gas for that area within the network. This will have an impact on avail- able resources and can impact the quantity of low-carbon gas entering the system from facilities such as biomethane plants and in the future perhaps hydrogen blending from power to gas facilities. When Bayesian statistics are applied to gas demand, probability distributions are generated for the amounts of gas used on future days. Why is this important? Because the future is unknown, and forecasts need to account for a range of potential sce- narios. This is where Bayesian methods have a major advantage to other statistical approaches, providing network operators with better predictive assurance than other approaches to gas demand forecasting.

• Cover