Flexbrick photovoltaic ceramic textile façade: a coupled EnergyPlus–Radiance–PV dataset and surrogate-optimisation pipeline across European climates

This deposit accompanies the study "Machine-Learning Surrogates and Simulation-Validated Multi-Objective Optimisation of a BIPV Ceramic Textile Façade: Energy, Daylight and Thermal Comfort Trade-Offs Across European Climates". It quantifies how the binary weave pattern of a 28 x 12 Flexbrick ceramic lattice - with each brick optionally hosting one photovoltaic cell - governs building energy use, daylight, thermal comfort and photovoltaic yield. The deposit contains the assembled main dataset (2,560 rows: 512 Sobol-sampled patterns across 5 European climates, 50 columns) in CSV and Parquet; a generalisation dataset of 1,950 cases spanning six architectural contexts (orientation east/west, room depth 3 and 7 m, window-to-wall ratio 30% and 75%); the geometry-only descriptor table; the bare-façade baseline KPIs; and a column-level data dictionary describing every variable with units, ranges and definitions. It also includes the base EnergyPlus building model, 65 trained surrogate models (random forest, XGBoost and MLP per indicator, plus context-aware "generalist" models), the full reproducible pipeline (EnergyPlus + Radiance + pvlib simulation, dataset assembly, surrogate training, SHAP interpretability, NSGA-III optimisation and validation), the EDA notebook, PNG previews of all 513 façade configurations, and the source tables of the paper's Supplementary Material (surrogate performance, context transfer, sensitivity studies, NSGA-III convergence, symmetry/order correlations, and a thermal-mass/vented-cavity spot-check of the massless-screen abstraction). Each pattern was evaluated, for every climate (IWEC typical years for Berlin, Barcelona, Seville, Helsinki and Turin), with coupled EnergyPlus 22.2 (energy and comfort), Radiance (climate-based daylight modelling) and pvlib (vertical south-facing BIPV) simulations on a single-zone 3 x 5 x 3 m perimeter office (15 m2 floor, south window-to-wall ratio 54%, ideal-loads HVAC). All performance indicators are anchored to a simulated bare-façade (no-screen) baseline. The surrogate models use a leakage-safe, configuration-grouped train/test split, and the NSGA-III optima are validated by full re-simulation. Energy-use intensities are reported per floor area (15 m2). The raw per-job simulation output (tens of GB) is not included but is fully reproducible from the released code; the five IWEC weather files are referenced by name in the README (not redistributed, ASHRAE copyright). Data, models, the data dictionary and the previews are released under CC BY 4.0; the source code is released under the MIT License. Please cite both this dataset (DOI above) and the accompanying paper.