Examples — the Kalobeyei Showcase¶
This section moves from theory to practice. Rather than looking at a single model run, it walks through a connected series of scenarios built on one real case study — the village of Kalobeyei in northern Kenya — and shows how the optimal energy system changes as we progressively add realism and planning complexity.
The goal is not only to read the final results, but to understand why different planning assumptions lead to different system designs.
What you will see
Starting from a simple off-grid hybrid mini-grid, we introduce — one layer at a time — a different storage technology, a physically-grounded battery ageing model, growing demand, staged (multi-step) investment, uncertainty about future grid connection, and finally an environmental externality (carbon cost). Each step reuses the previous set-up and changes only what the lesson is about.
The scenario roadmap¶
1. Baseline off-grid hybrid: solar PV + diesel + lead-acid battery, constant demand
│ swap the storage technology
2. Lithium-ion same system, lithium-ion instead of lead-acid
│ replace the assumed ageing rate with a physical model
3. Battery degradation semi-empirical cycle + calendar fade, driven by site temperature
│ let demand grow
4. Demand growth +5 %/year demand over the 10-year horizon
│ allow phased investment
5. Capacity expansion the same growth, built in two investment steps
│ add uncertainty
6. Stochastic grid two possible futures for when (and how reliably) the grid arrives
│ price emissions
7. Carbon cost the same futures, with CO₂ costs inside the objective
Results at a glance¶
The seven scenarios share the same site, demand shape, and techno-economic assumptions, so their headline results are directly comparable.
Expected Net Present Cost (NPC) and Levelized Cost of Electricity (LCOE) for the seven scenarios.
Final installed capacity by technology. Demand growth enlarges every component; grid access and carbon pricing shift the mix toward renewables.
| # | Scenario | NPC (kUSD) | LCOE (USD/kWh) | Solar PV (kW) | Battery (kWh) | Diesel (kW) | Renewable share |
|---|---|---|---|---|---|---|---|
| 1 | Baseline — lead-acid | 618.0 | 0.223 | 302 | 1332 | 26 | 91.6 % |
| 2 | Lithium-ion | 526.6 | 0.190 | 281 | 795 | 26 | 91.2 % |
| 3 | Battery degradation | 519.0 | 0.188 | 271 | 782 | 27 | 90.5 % |
| 4 | Demand growth | 680.0 | 0.201 | 329 | 897 | 49 | 85.1 % |
| 5 | Capacity expansion | 663.9 | 0.196 | 386 | 1087 | 45 | 88.9 % |
| 6 | Stochastic grid | 602.0 | 0.178 | 285 | 780 | 39 | 78.4 % |
| 7 | Carbon cost | 638.4 | 0.189 | 300 | 825 | 38 | 81.1 % |
All monetary values are expected (probability-weighted) present values in real USD. Capacities are end-of-horizon values. Renewable share is the horizon renewable contribution; grid imports count as non-renewable.
How to read this section¶
The examples are grouped into three short chapters, each pairing two scenarios:
- The Kalobeyei Case Study — the community, and the demand and resource inputs shared by every scenario.
- Technology Choice — baseline lead-acid vs. lithium-ion, why a cheaper component does not mean a cheaper system, and what changes when battery ageing is modelled physically instead of assumed.
- Planning for Growth — demand growth, and staged capacity expansion as an adaptive response.
- Uncertainty & Externalities — stochastic grid connection and carbon pricing.
Reproducibility¶
The seven scenarios are published as a citable dataset rather than bundled with the source code — together they are several hundred megabytes, which does not belong in a repository people clone to install the package:
Kalobeyei case-study dataset
Each scenario ships as a complete project folder: inputs/ (the data needed to re-solve
it), results/ (the tables every figure on these pages is drawn from) and logs/ (the
solver log of the published run).
Every number in this section is read directly from those results/ folders, so the figures
and the archive cannot drift apart.
| Scenario | Project folder in the archive |
|---|---|
| 1 — Baseline (lead-acid) | Kalobeyei_1 |
| 2 — Lithium-ion | Kalobeyei_2 |
| 3 — Battery degradation | Kalobeyei_3 |
| 4 — Demand growth | Kalobeyei_4 |
| 5 — Capacity expansion | Kalobeyei_5 |
| 6 — Stochastic grid connection | Kalobeyei_6 |
| 7 — Carbon cost | Kalobeyei_7 |
To re-solve a scenario, download the archive and unpack the project folders into a
projects/ directory inside your workspace:
Then point MicroGridsPy at that workspace and solve:
import microgridspy as mgp
mgp.set_workspace("my-workspace") # or run from inside it
model = mgp.solve("Kalobeyei_2", solver="highs") # lithium-ion reference
results = model.results()
print(results.kpis)
The published runs used Gurobi 13.0.3; HiGHS reproduces the same optima. Solver versions can shift the last digits of a degenerate LP, so expect agreement to a few decimal places rather than bit-for-bit equality.
Solving scenario 3
The semi-empirical degradation model discretises the battery into state-of-charge bands,
which makes scenario 3 roughly four times larger than the others (3.2 M rows against
0.9 M) and much slower to solve. The published run used Gurobi with Crossover=0:
barrier converges in about four minutes, whereas the crossover to a vertex solution ran
for over an hour without changing the objective.