Python simulation framework for 15-year techno-economic comparison of PEM and Alkaline water electrolysers under real-world renewable energy conditions.
Master Thesis — Shubham Manchanda | Hydrogen Technology & Economics, TH Ingolstadt | Conducted at Fraunhofer IFF, Magdeburg
Green hydrogen is central to Europe's energy transition, but the investment decision between PEM and Alkaline electrolysers remains unclear under variable renewable supply. This model quantifies that decision with real engineering data — making it directly useful for energy companies evaluating electrolyser projects.
| Metric | PEM (35 MW, 12 t storage) | Alkaline (20 MW, 10 t storage) |
|---|---|---|
| LCOH | €9.06 /kg H₂ | €4.86 /kg H₂ |
| Demand Satisfaction | 95.4% | 96.0% |
| Capacity Factor | 41.1% | 73.0% |
| 15-Year NPV | −€62.5M | +€18.2M |
| Stack Replacements | 1 (at ~60,000 h) | 1 (at ~80,000 h) |
| Monte Carlo LCOH Range | €6.24 – €12.64 /kg | €4.35 – €7.04 /kg |
Bottom line: Alkaline achieves 46% lower hydrogen cost than PEM under identical renewable conditions, primarily due to lower capital cost and longer stack lifetime.
- Electricity cost dominates LCOH — renewable generation cost (LCOE) accounts for 50–65% of total hydrogen cost for both technologies, making it the single most important lever for project economics
- Alkaline wins on cost, PEM wins on flexibility — Alkaline's lower CAPEX (€1,150 vs €1,950/kW) and longer stack lifetime (80,000 vs 60,000 h) drive its cost advantage, but PEM handles variable renewable input with wider load range (5–100% vs 20–100%)
- Degradation increases LCOH by 8–12% over 15 years — voltage degradation (1.5 µV/h PEM, 0.7 µV/h Alkaline) progressively reduces efficiency, requiring stack replacement that adds €0.30–0.80/kg to lifetime cost
- System sizing is non-trivial — undersized electrolysers waste renewable energy through curtailment; oversized systems increase CAPEX without proportional output gains. The optimum is configuration-specific, not rule-of-thumb
- Uncertainty is significant — Monte Carlo analysis shows LCOH can vary by ±40% (PEM) and ±30% (Alkaline) depending on input assumptions, highlighting the risk of deterministic-only analysis
Figure 1 — Side-by-side comparison of PEM and Alkaline electrolyser performance across key metrics
Figure 2 — LCOH waterfall breakdown showing electricity cost as the dominant driver
Figure 3 — Tornado sensitivity analysis: electricity price dominates LCOH uncertainty for both technologies
Figure 4 — Monte Carlo simulation (N=10,000) showing LCOH probability distributions
- Hourly simulation — 131,400 timesteps over 15 years with real wind+solar profiles (132 MW installed capacity)
- Electrochemical modelling — Nernst-based reversible voltage, Butler-Volmer activation, ohmic and concentration overpotentials
- Degradation model — Voltage degradation (µV/h) with stack replacement logic based on efficiency thresholds
- Grid-search optimisation — 600 configurations (10 electrolyser sizes × 10 storage capacities × 6 renewable fractions)
- Economic analysis — LCOH, NPV, IRR, payback period, CAPEX/OPEX breakdown with oxygen revenue credits
- Uncertainty quantification — Monte Carlo simulation (N=10,000) + tornado sensitivity charts
- Publication-ready plots — 90+ figures generated automatically in PNG and PDF
Input Data Simulation Engine Economic Output
────────── ───────────────── ───────────────
Wind + Solar power → Electrolyser model → LCOH (€/kg H₂)
(8,760 h/year × 15yr) • Polarisation curve NPV, IRR
• Degradation tracking Payback period
H₂ demand profile → • Stack replacements → Cost breakdown
(hourly kg) • Storage buffer dynamics (CAPEX, OPEX, energy)
Optimisation: grid search over 600 configurations
Selection: lowest LCOH meeting ≥95% demand satisfaction
Cell voltage model:
V = E_rev(T, P) + η_activation + η_ohmic + η_concentration
# Clone and set up
git clone https://github.com/shubham0429/Hydrogen-Electrolyser-Optimisation.git
cd Hydrogen-Electrolyser-Optimisation
python3 -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
# Run PEM 15-year simulation (~6 min)
cd source_code
python run_pem_thesis_final.py
# Run Alkaline 15-year simulation (~4 min)
python run_alkaline_thesis_final.pyResults appear in results/pem_thesis_final/ and results/alkaline_thesis_final/.
| Script | Purpose | Runtime |
|---|---|---|
run_pem_thesis_final.py |
PEM 15-year simulation + 8 figures | ~6 min |
run_alkaline_thesis_final.py |
Alkaline 15-year simulation + 8 figures | ~4 min |
corrected_thesis_plots.py |
Comparison plots + optimisation heatmaps | ~30 sec |
sensitivity_analysis_alkaline.py |
Monte Carlo (N=10,000) + tornado charts | ~50 sec |
pem_optimization_v3.py |
PEM grid search (600 configs × 6 RE)¹ | ~3–5 hrs |
alkaline_optimization_v3.py |
Alkaline grid search¹ | ~3–5 hrs |
¹ Pre-computed results are included in
results/data/. Only re-run if changing parameters.
├── README.md
├── ABSTRACT.txt Thesis abstract (200 words)
├── requirements.txt numpy, scipy, pandas, matplotlib
│
├── data/
│ ├── combined_wind_pv_DATA.mat Hourly wind+solar power (8,760 h)
│ └── Company_2_hourly_gas_demand.csv Hourly hydrogen demand profile
│
├── source_code/
│ ├── sim_concise.py PEM electrolyser model (core engine)
│ ├── sim_alkaline.py Alkaline electrolyser model (core engine)
│ ├── electrochemistry.py Electrochemical equations
│ ├── data_loader.py MATLAB .mat file loading
│ ├── run_pem_thesis_final.py → Run this for PEM results
│ ├── run_alkaline_thesis_final.py → Run this for Alkaline results
│ ├── pem_optimization_v3.py Grid search optimisation (PEM)
│ ├── alkaline_optimization_v3.py Grid search optimisation (Alkaline)
│ ├── corrected_thesis_plots.py → Comparison figures (recommended)
│ ├── sensitivity_analysis_alkaline.py Monte Carlo + sensitivity
│ └── ... Additional plotting scripts
│
└── results/
├── pem_thesis_final/ PEM: 131,400-row timeseries + 8 figures
├── alkaline_thesis_final/ Alkaline: same structure
├── thesis_final_plots/ Best figures for publication
├── plots/ All 90+ thesis figures
└── data/ Pre-computed optimisation CSVs
├── pem_grid_search_all_RE.csv 416 PEM configurations
├── alkaline_grid_search_all_RE.csv 600 Alkaline configurations
├── monte_carlo_results.csv
└── sensitivity_results.csv
| Language | Python 3.9+ |
| Dependencies | numpy, scipy, pandas, matplotlib |
| Input data | 132 MW wind+solar (MATLAB .mat), hourly H₂ demand (CSV) |
| Simulation horizon | 15 years (131,400 hours) |
| Optimisation | Exhaustive grid search, 3-criteria hierarchical selection |
| Uncertainty | Monte Carlo N=10,000, seeded (seed=42) for reproducibility |
| Tested on | macOS (Python 3.9.6), compatible with Linux/Windows |
| Problem | Solution |
|---|---|
ModuleNotFoundError |
Run pip install -r requirements.txt inside activated venv |
FileNotFoundError: .mat file |
Run scripts from source_code/, not the root folder |
python: command not found |
Use python3 instead |
| Optimisation takes hours | Normal. Use pre-computed CSVs in results/data/ |
Shubham Manchanda
Hydrogen Technology & Economics (core modules completed), TH Ingolstadt
Thesis conducted at Fraunhofer IFF, Magdeburg
This project is licensed under the MIT License.
Feel free to use, adapt, or build on this work with attribution.