zceceal/zambia-explicit-peak

Explicit peak demand representation in least-cost electrification modelling: evidence from Zambia # Explicit peak demand in least-cost electrification modelling — Zambia A controlled OnSSET experiment testing whether representing **peak demand per settlement**, rather than through one national load factor, changes the least-cost electrification plan for Zambia. **Headline: doing so raises the modelled lifetime cost of universal access by +49.9%, and changes the least-cost technology for 34,461 settlements (12.7%) — see Headline results below, or `results/summary/` for the committed numbers behind every figure, with no setup required.** Model and code behind the MSc research paper *Explicit peak demand representation in least-cost electrification modelling: evidence from Zambia* (Imperial College London, 2026). --- ## The question, in one paragraph Geospatial electrification models decide, for every settlement in a country, whether extending the grid, building a mini-grid, or installing stand-alone solar is cheapest. To do that they need each settlement's **peak** power, but the standard workflow specifies only **annual energy** and converts it to a peak with a single national factor — the same factor for a ten-household hamlet and a three-thousand-household town. That is physically wrong: in a small settlement everyone cooks and lights at once, while in a large one demands average out. This repository replaces that single factor with a peak computed per settlement from its connection count, holds everything else identical, and measures what changes. ## The experiment | Arm | Name | Peak demand representation | |-----|------|----------------------------| | **R0** | Energy-only baseline | One uniform load factor for every settlement (standard practice) | | **R1** | Explicit peak | Peak-to-energy ratio computed per settlement from its connection count | Same 270,526-settlement spine, same costs, same resource data, same solver, same random seed. The demand pre-processor is the only difference, which is what makes the comparison controlled. ## Headline res …