GP-SA-MOEA reproducibility package: code, results, derived data, and raw third-party inputs
Year: 2026
Type: article
Abstract: Reproducibility package (v2, revised manuscript) for the GP-SA-MOEA project - a Gaussian-process-surrogate-assisted multi-objective evolutionary algorithm for national-scale road-network trip-time prediction and EV charging-station placement, evaluated against an exact multi-source Dijkstra baseline. Companion package of the manuscript "Exact Multi-Source Evaluation for National-Scale EV Charging Station Placement: How Far Does Surrogate Assistance Take Us?" (revision submitted to ENERGIES, MDPI). Split into three archives: GP-SA-MOEA_code_results_20261011.zip (363 MB) - full code (src/, experiments/, tests/), London ground-truth fronts and per-seed run outputs (results_b/), national-scale per-point outputs and merged reference front (results_c/), Figure-1 ground-truth arrays (results_d/), complete configurations, seeds, and dataset SHA-256 hashes (config_e/), preprocessing entry points (scripts_f/), and all regenerated figures incl. the Graphical Abstract (figs/). GP-SA-MOEA_derived_data_20261011.zip (1.49 GB) - pipeline outputs that are the runtime inputs of all reported runs: 12.8M-node weighted road graph (road_graph.pkl), node features/coords/regions, 5.4M demand points with weights, GP train/val/test splits, and NUTS1 boundaries; regenerated for the revised manuscript, including the rebuilt London network (1,087,873 nodes / 2,049,648 directed edges after the one-way edge correction) and the exact eight-point reference front. Regenerable from the raw inputs via data_pipeline.py. GP-SA-MOEA_raw_inputs_20260830.zip (1.77 GB) - original third-party inputs, unchanged from v1: Geofabrik England OSM extract england-260728.osm.pbf (ODbL 1.0), WorldPop gbr_ppp_2020_UNadj.tif (CC-BY 4.0), and ONS NUTS Level 1 boundaries (OGL v3.0). Each file keeps its upstream license. Changes in v2 (manuscript revision R1): the Group E feature-set ablation was re-run through the shared GP-SA-MOEA runner after the one-way-edge implementation fix, on the rebuilt London network with exact Dijkstra re-evaluations (three seeds per configuration); national-scale K-threshold sensitivity analysis; exact re-evaluations of all method fronts and (f2, f4)-knee design-point sensitivity; surrogate calibration validated on a fixed hold-out set. Code, results, and derived data were regenerated accordingly; raw third-party inputs are byte-identical to v1. Reviewer verification pointers (details in the in-archive README.md): per-seed ablation outputs in results_b/, national exact re-evaluation arrays and merged reference front in results_c/, Figure-1 plotted arrays in results_d/fig1_bin_stats.csv, and the national sensitivity sweep in results_c/exp7_national_sensitivity.json.
Institutions: University of Bristol, Hubei Polytechnic University
Language: English
Subfield: Computational Theory and Mathematics
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7204660087