# Dataset for: Degradation-aware tri-objective optimisation of gigawatt-scale PEM green hydrogen production: the hidden cost of ignoring stack degradation

**DOI:** https://doi.org/10.5281/zenodo.21414621
**Authors:** Majed Alanazi (University of Sheffield, ORCID 0009-0000-7327-0309), Kevin J. Hughes (University of Sheffield), Mohamed Pourkashanian (University of Sheffield)
**Licence:** Creative Commons Attribution 4.0 International (CC BY 4.0)
**Version:** 1.0

## Overview

Results data supporting the manuscript *"Degradation-aware tri-objective
optimisation of gigawatt-scale PEM green hydrogen production: the hidden cost
of ignoring stack degradation"*, submitted to Applied Energy. The dataset
contains: (1) the canonical Pareto-optimal design vectors from the
tri-objective NSGA-III optimisation of a gigawatt-scale PEM green hydrogen
facility at NEOM, Saudi Arabia; (2) the twenty-run statistical
reproducibility campaign; (3) inter-annual resource-robustness evaluations
across eighteen NSRDB solar years (2005-2022) and twenty-one ERA5 wind years
(2004-2024); and (4) sequential 25-year degradation-rollout trajectories for
the four canonical designs.

The simulation model and optimisation code are available from the
corresponding author on reasonable request.

## How to cite

Please cite both the dataset (via the DOI above) and the associated article.

## Files

| File | Contents | Rows |
|---|---|---|
| 01_canonical_designs.csv | Four canonical designs: full decision vector and headline outcomes | 4 |
| 02_campaign_20run_summary.csv | Twenty independent NSGA-III runs (seeds 1-20): minima and minimum-cost architectures | 20 |
| 03_robustness_solar_18yr.csv | Canonical designs re-evaluated over 18 NSRDB solar years (2005-2022), wind fixed | 72 |
| 04_robustness_wind_21yr.csv | Canonical designs re-evaluated over 21 ERA5 wind years (2004-2024), solar fixed | 84 |
| 05_rollout_summary.csv | Sequential 25-year rollout vs single-year analytic levelised cost, per design | 4 |
| 06-09_rollout_trajectory_*.csv | Year-by-year 25-year trajectories for each canonical design | 25 each |

## Column dictionaries

### 01_canonical_designs.csv
- `design` - canonical design label (Best-LCOH, Knee, Max-H2, Min-degr)
- `CSP_scale` / `PV_scale` / `Wind_scale` - dimensionless capacity multipliers
  applied to base reference capacities of 130.6 MW (CSP+TES), 2,000 MW-DC
  (PV) and 166.9 MW (wind) respectively; e.g. PV_scale x 2,000 MW-DC gives
  installed PV capacity
- `N_stacks` - number of PEM electrolyser stacks (50 cells, 280 cm2 active
  area each)
- `E_batt_MWh` - battery energy capacity, MWh
- `min_load_fraction` - minimum stack load fraction (-)
- `replacement_threshold` - cumulative voltage-degradation fraction
  triggering stack replacement (-)
- `N_compression_stages` - hydrogen compression stages (integer)
- `j_nom_A_per_cm2` - nominal current density, A cm-2
- `j_max_ratio` / `j_min_ratio` - overload and turndown ratios relative to
  j_nom (-)
- `LCOH_USD_per_kg` - levelised cost of hydrogen, single-year basis, USD/kg
- `H2_tonnes_per_year` - annual hydrogen production, tonnes
- `degradation_mV_per_year` - fleet-mean voltage degradation rate, mV/yr
- `capacity_factor` - fleet capacity factor referenced to rating at maximum
  permitted current density (-)

### 02_campaign_20run_summary.csv
- `run` - random seed / run index (1-20)
- `LCOH_final_archive_USD_per_kg` - minimum LCOH retained in the final
  archive after reference-direction niching, USD/kg
- `LCOH_discovered_USD_per_kg` - minimum LCOH discovered anywhere in the
  search history, USD/kg
- `CSP_capacity_GW` / `PV_capacity_GW_DC` / `Wind_capacity_GW` - installed
  capacities of that run's minimum-cost design
- `N_stacks`, `E_batt_GWh`, `j_nom_A_per_cm2` - as above (battery here in GWh)

### 03_robustness_solar_18yr.csv
- `design` - canonical design label
- `meteorological_year` - NSRDB solar year used for the PV/CSP profiles
  (2005-2022); wind profile fixed at reference
- `LCOH_USD_per_kg`, `H2_tonnes_per_year`, `capacity_factor` - outcomes for
  that design under that solar year

### 04_robustness_wind_21yr.csv
- `design` - canonical design label
- `meteorological_year` - ERA5 wind year (2004-2024); solar fixed at
  reference
- `LCOH_USD_per_kg`, `H2_tonnes_per_year`,
  `degradation_mV_per_year`, `capacity_factor` - outcomes for that design
  under that wind year
- `feasible` - 1 if all operating constraints satisfied

### 05_rollout_summary.csv
- `design` - canonical design label
- `LCOH_analytic_USD_per_kg` - conventional single-year levelised cost
- `LCOH_rolled_USD_per_kg` - levelised cost from the 25-year sequential
  rollout (cohort degradation state carried year to year)
- `delta_USD_per_kg`, `delta_percent` - rolled minus analytic
- `H2_year1_tonnes`, `H2_year25_tonnes` - annual production in project years
  1 and 25
- `replacement_events_total` - stack-cohort replacement events over 25 years

### 06-09_rollout_trajectory_[design].csv
- `project_year` - 1 to 25
- `H2_tonnes` - hydrogen produced that year
- `fleet_voltage_degradation_mV` - cumulative fleet-mean voltage degradation
  at year end, mV
- `specific_energy_consumption_kWh_per_kg` - system specific energy
  consumption that year
- `mean_loading_relative_to_nominal` - fleet mean loading relative to
  nominal rating
- `replacement_events` - cohort replacement events that year
- `replacement_cash_MUSD` - replacement expenditure that year, million USD

## Provenance

Generated by the study's MATLAB simulation and optimisation pipeline
(hourly 8,760-h dispatch with ten-cohort degradation tracking; NSGA-III,
population 300, 200 generations). Renewable input profiles: NREL System
Advisor Model 2025.4.16 (NSRDB satellite irradiance, site 28.21 N 34.90 E,
meteorological year 2005) and a 21-year ERA5-based wind record. Economic
parameters anchored to realised Saudi project costs as described in the
article. Column headers in this deposit were expanded from the pipeline's
native short names for readability; values are unmodified pipeline output
(correction-set of 2026-07-02).

Native-to-deposit header mapping is one-to-one and documented above; the
underlying numerical content is byte-identical to the pipeline output apart
from line-ending normalisation.

## Integrity

SHA-256 checksums for every file are listed in `SHA256SUMS.txt`.
Verify with:  `shasum -a 256 -c SHA256SUMS.txt`