Forecasting peak gas demand for the UK Future Energy Scenarios

The Smith Institute independently reviewed the technical basis of the 1 in 20 peak demand forecasting methodology for National Grid.

The problem

Every year at National Grid, the National Transmission System (NTS) team provides the Network Strategy team with the year-ahead annual gas demand forecasts. These forecasts are used by the Network Strategy team to book gas supply capacity to meet the UK’s year-ahead gas demand for the whole year. Understanding the accuracy and robustness of these demand forecasts is critical to the Network Strategy team’s business.

The Smith Institute was asked by National Grid to conduct an independent review of the technical basis of the 1 in 20 peak demand forecasting, and so underpin the confidence of the Network Strategy team in the use of such forecasts for booking year-ahead gas capacity.

The Smith Institute found that the methodology used to generate the 1 in 20 peak demand forecast is fit for purpose, and suggested enhancements to improve its performance.

The solution

To assess the methodology used for generating the 1 in 20 peak demand forecast and to identify potential enhancements to the existing methodology, the Smith Institute independently developed mathematical and statistical models and applied them to data provided by the NTS and Network Strategy teams. Outcomes from the modelling and analysis were regularly reported to and discussed with both teams to gain shared insights.

Our analysis comprised the following key stages:

  • reviewing and scrutinizing each step in the documented methodology used to generate peak gas demand forecasts;
  • developing models for demand forecasting that make effective use of raw historical data;
  • testing relationships between variables in the existing methodology;
  • validating data-related and operational assumptions in the existing methodology; and
  • developing an understanding of why recent demand forecasts from NTS have been perceived to be high by the Network Strategy team.

To conclude the project, we held a workshop at National Grid to present our findings, to discuss the feasibility of their application and to provide the opportunity for feedback from National Grid. During the workshop, we highlighted aspects of the methodology that need improving, and categorised our recommendations into those for immediate improvement to the 1 in 20 peak estimates and those for implementation in the medium to long-term as part of planned improvements to National Grid’s demand forecasting methodology. Interesting discussion arose over the validity of the assumption that National Grid should always base their forecasts on the 1 in 20 peak, particularly as we had demonstrated the sensitivity of peak estimates to composite weather variable history: a longer history results in higher peaks.

The benefit

Both the NTS and Network Strategy teams found the analysis valuable in evaluating and further underpinning the robustness of the process used by National Grid.

The causal links between variables that shape peak demand are shown in the following picture, which illustrates the stages in the demand forecast methodology used by National Grid.


The Smith Institute’s key findings can be summarised as follows:

  • different loads bands respond in different ways to economic growth scenarios. For example, in the Going Green scenario domestic load is particularly responsive, whereas in the Slow Progression economic growth scenario the demand in the SME segment is significantly affected;
  • peak demand forecasts are significantly influenced by economic assumptions within future energy scenarios for gas; and
  • Composite Weather Variable (CWV) history has a significant impact on peak demand estimates: specifically, longer history results in higher peaks.

The Smith Institute is confident that the methodology used to generate the 1 in 20 peak demand forecast is fit for purpose, however improvements can be made to enhance its performance. In particular, we were able to suggest:

  • improvements to the modelling of the peak demand distributions;
  • improvements to the modelling of the random error term in the demand versus CWV regression model; and
  • a novel formulation of the demand versus CWV relationship based on support vector regression.

Since the completion of our analysis, the NTS team has implemented the suggested improvements to the random error term for the CWV regression model. NTS is also planning to include the Smith Institute's recommendations into National Grid's changes for next year’s demand forecast methodology, in preparation for creating next year’s year-ahead forecasts. Our comparison between the current (Gumbel) distribution and our proposed (Weibull) distribution, for estimating the likelihood of different peaks, showed that the proposed distribution yielded higher confidence in the forecasts of 1 in 20 peaks over longer horizons. NTS intends to capitalise on these findings.

The project was funded via Network Innovation Allowance funds.

As an outcome this study provided assurance that our approach remains appropriate and low risk. We intend to make use of this vital report in reviewing the adoption of a shorter historical data sequence. We are obliged to include at least 50 years of historical data in compliance with our regulatory licence and intend to adopt an updated climate change adjusted weather history as the basis of our method from 2015. Furthermore we intend to review the most appropriate distribution curves as part of a model refresh. The review by the Smith Institute was both rigorous and professional, providing both assurance and furthering learning for our business. The study has provided a timely input and supportive of our needs and the Smith Institute were a pleasure to work with.

Dr Stephen MarlandGas Demand Manager at National Grid
Now Recruiting