Learnings

Outcomes

The project aimed to improve NESO’s understanding of the long-term decline in reactive power demand from distribution networks and to establish a practical basis for projecting future reactive power behaviour at GSP level.  

The first outcome, delivered through WP1, was determining the causes of declining reactive power demand. The review confirmed that the trend is not driven by a single factor, but by a combination of increased underground cabling, lower net active demand, growth in embedded generation, changing demand-side technologies and network operating practices. The analysis of NESO data also showed a clear long-term trend towards more negative reactive power values and more frequent reactive power export from distribution networks to the transmission system, particularly during low-demand periods.  
 
The second outcome, delivered through WP2, was the development of a methodology for producing long-term reactive power projections. WP2 concluded that simple historical extrapolation or fixed power factor assumptions would not be sufficient, as reactive power behaviour varies materially between GSPs and is strongly affected by network topology, embedded generation, demand characteristics and voltage control behaviour. A hybrid methodology was therefore defined, combining GSP clustering, representative network archetypes, network simulation, statistical emulation and national scaling. This allows future reactive power behaviour to be assessed over a 15-year horizon under different demand, generation and network development scenarios, aligned with FES and DNO planning inputs.  
 
The third outcome, delivered through WP3, was the implementation of this methodology into a working model framework. A modular Python-based codebase was developed, covering feature engineering, GSP clustering, active learning, GAM-based emulation, uncertainty quantification and Power Factory automation where suitable network models are available. Four archetype GSPs were selected and developed for the initial model implementation: Norton, Macclesfield, Stirling and City Road. The model structure also includes the interfaces required for Stage 3 national scaling and projection, which will be completed and exercised in WP4.  
 
The project has also improved NESO’s capability to undertake future analysis. WP1 brought together historical active and reactive power data for 340 GSPs at half-hourly resolution. WP2 converted the learning from that data and from network sensitivity studies into a practical projection methodology, and WP3 moved the work from methodology design into an implemented modelling framework. This represents a clear progression from evidence gathering to a model structure that can support future national projection studies. Full national projection runs have not yet been completed, as these form part of the next stage of the project. However, the project has progressed from review and methodology definition into an implemented modelling framework, with a clear route to national application, tool deployment and future integration into business-as-usual planning processes through the later work packages.  

Lessons Learnt

One of the main lessons from the project so far is that reactive power behaviour at the transmission-distribution interface cannot be treated as a simple extension of active power behaviour. The analysis showed weak and highly variable relationships between active and reactive power at many sites, with reactive power influenced by a wider set of factors including cable penetration, embedded generation, demand-side technology changes, voltage control behaviour and network topology. Future projects should consider multi variable studies and mut take into account the need of several assumptions for long term projections. 
 
The work also highlighted the importance of data quality and consistency. NESO data provides a strong national basis for identifying long-term trends, but the comparison with DNO data showed differences in coverage, measurement points, sign conventions and agreement between datasets. These issues do not prevent useful analysis, but they do need to be identified early and managed carefully. Future projects would benefit from agreeing common data definitions, measurement points and sign conventions at the start, particularly where outputs are intended to support national-scale modelling.  
 
Stakeholder engagement also proved important. Feedback from DNOs and TOs helped confirm that current reactive power forecasting practices often rely on historical power factor assumptions, and that there are important operational and planning factors that are not always visible in public datasets. Future work should continue engaging with DNOs and TOs, particularly around network development plans, voltage control settings, DER connection requirements and the availability of more detailed network data.  
 
A final learning is that the model needs to remain flexible. The WP3 implementation has been developed in a modular way, with separate components for clustering, emulation, projection, PowerFactory automation and data processing. This is important because input data, FES assumptions, DNO network data and modelling requirements are likely to evolve. A modular structure should make it easier to update the tool and incorporate improved datasets or methods in future phases.