Learnings

Outcomes

The work in this project set out to develop approaches to probabilistic projections that are complementary to existing probabilistic methods for planning activities carried out by National Gas. This has been achieved using a combination of state-of-the-art machine learning and statistical techniques, including machine learning models Gradient Boosted Machines (GBMs) and Generalised Additive Models (GAMs), along with weather rescaling processes, and climate corrections, to demonstrate the value of the projections through a range of use cases reflecting different decision making processes on planning timescales. Importantly, the approach developed here is consistent with the methods used in the Short-Term Forecasting project, delivered by TNEI for National Gas, providing a consistent framework across all time horizons for operational and planning activities. The Projections project has also demonstrated an approach to account for the near-term impacts of climate change, proving a comparison with results which assume weather patterns remain unchanged. This is an important step towards alignment with UKCP18 climate projections and provides the foundations for National Gas to incorporate climate change considerations into planning decision making processes. Finally, many of the supporting processes developed in this work may be useful beyond the demonstration of the use cases presented here, building foundations for adopting these methods into existing business processes.

Project outcomes include customisable code with notes available that demonstrates the following for the specific use cases: 

•        Scaling of weather-rescaled time series to the target scenario year by matching annual values in the FES; 

•        Calculation of gas power station demand from relevant electricity components, assuming constant nuclear generation and an efficiency conversion to gas demand; 

•        Summing of gas consumption demand components to calculate time series of projected total daily gas demand and aggregation to the winter season total. 

•        Applying the individual annual projections of consumption demand to the model that describes NTS capability to produce projections of that NTS capability. 

•        Combining the NTS capability projections with projections of LDZ demand within the zone to derive estimates of the capacity to absorb supply within the zone. 

•        Comparing this capacity to scenarios of LNG supply to identify when that supply would be limited.

Value tracking                                                Data Point                                       Data Point Definition

Maturity                        TRL 4-5                                      Initially at concept level (NIA Project) but has matured to 4-5 with development of actionable code. 

Opportunity                       100% of single asset class                Models can be applied across the business to project LDZ supply and demand of gas with notebooks readily available  

Deployment costs              £0                                                      Unknown but minimal deployment costs due to software nature 

Innovation cost                  £162,667                                           Cost of innovation project

Financial Saving                 £0                                                      Unknown, research project 

Safety                                 0                                                        Not project focus 

Environment                      0.0                                                     No environmental impact  

Compliance                                                                                  Supports compliance 

Skills & Competencies                                                                 No change 

Future proof                      Supports business strategy              High quality forecasts/projections are critical for the safe, secure, and efficient planning and operation of Great Britain’s gas system, informing decisions on multiple time scales.

Lessons Learnt

Lessons learnt for future projects include: 

•        Incorporating climatology models into supply/demand projections can be complex, always include contingency time in project plans for projects involving the models in future. 

•        Data gathering can be time consuming and require stringent data access credentials depending on its source. Ensure risk mitigation strategies are robust enough to reduce delays in gathering data. 

•        Legal do not allow NGT as a business to share information via GitHub easily therefore secure access via SharePoint is the best way to manage data access for sharing/modifying code.

•        Ensure Christmas holidays are incorporated in project Gannt chart in future to avoid delays in project execution.

•        Ensure simplified and less technical jargon is used in executive summary and final report to improve readability and accessibility to those other than SMEs of data science.

•        Engaging with academic partners can introduce innovative approaches to our projects through their specialised methods. These partners may also have access to proprietary datasets or supercomputers that we do not, which could result in delays in acquiring, processing, and replicating these methods. It is essential to set clear goals at the outset, identify potential obstacles, and develop an action plan to address them.

•        We recommend establishing a more robust review process using approaches such as Git collaboration. This method allows the supplier to break down large tasks into smaller, reviewable pieces, with comprehensive commentary tracking. By capturing changes and addressing questions or issues early, we can better adhere to the program timeline and ensure timely finalisation of deliverables. It is crucial to agree on a collaboration system upfront to facilitate this process.