Issue link: https://fhpublishing.uberflip.com/i/1078368
GAS Counting on gas Providing accurate forecasts for gas demand and maintaining systems balance can be challenging. A new approach aims to address this. Matt Linsley, Keith Owen and Brett Cherry explain. A team of statistical research- ers in the School of Maths, Statistics & Physics at New- castle University and analysts at Northern Gas Networks (NGN) have developed and implemented Bayesian statistics predictive models for energy demand. NGN deliver gas to 2.7 million homes and businesses in the North East, Northern Cumbria and much of Yorkshire. NGN use the real-time computational methods within their distribution system operator function to accurately forecast gas demand across different timescales, at regional and local levels, thereby maintaining system balance, enhancing stock utilisation and increasing network efficiency. Efficient and high-quality forecasting for short and long-term gas demand enables the gas industry to optimise the network for efficiency, understand uncertainty in their forecasts, and mitigate risk. Bayesian methods incorporate key factors that are (in general) omitted from other quantitative sta - tistical models, such as process knowledge, domain expertise and experience. Incor- porating these factors within the statistical analysis approach, in conjunction with tra- ditional statistical inference, creates more accurate forecasts; improving operational efficiency, situational awareness and the overall functionality of the physical system. The approach to demand forecasting developed by Newcastle University inte - grates a combination of expert knowledge and data analysis to generate predictions. The method produces probability based predictions which better capture the uncer- tainty in future demand, whether this be an hourly, daily, monthly or annual prediction. A unique attribute of the Bayesian forecasts allows for real-time updating in the fore- casts as recent data is captured. It has been recognised by the industry that while distribution networks all have NETWORK / 27 / FEBRUARY 2019