Reactive Power Demand Projection from Distribution Networks in Great Britain

This paper presents a hybrid methodology combining physics-based network modelling with statistical emulation using Generalised Additive Models (GAMs). Empirical analysis establishes that installed distributed resources capacity explains reactive power behaviour more effectively than instantaneous generation output, and that the relationship between active and

Q is highly site-specific and spans a significant range, precluding uniform assumptions.

Network sensitivity studies using archetypal distribution models can help quantify how Q changes propagate differently through various topologies. A case study implementation for an urban distribution network validates the methodology. An active learning framework achieves over 90% reduction in simulation burden while maintaining emulator accuracy (R² = 0.98). Training the emulator on physics-based simulation outputs rather than measured data (R² = 0.37) provides a more suitable structure for projection under future scenarios not represented in historical observations.

The approach, while early in its development, shows strong potential to enable scenario-based assessment of future Q behaviour under different pathways of network evolution and demand and distributed generation growth, helping to provide system operators and network planners with a valuable tool and approach for long-term voltage related investment planning.