Data and code for target-dependent cleaner production screening of recycling pathways for end-of-life electric-vehicle lithium-ion batteries

Description

This dataset contains the curated data and reproducible Python code supporting the study “Target-dependent cleaner production screening of recycling pathways for end-of-life electric-vehicle lithium-ion batteries using an interpretable machine-learning framework”. Version 2 retains the Version 1 analysis paths and adds physical-configuration-grouped evaluation for all six screening targets on cell and pack bases, route-held-out transfer stress tests, target-specific feature exclusions, dependence-aware uncertainty analyses, active-cathode spend-proxy sensitivity outputs, and machine-readable validation records. The package includes the 1,008-row designed scenario grid; deterministic cell- and pack-basis screening targets; the prespecified 168-row machine-learning input set representing 84 paired physical configurations; seven-model comparison results; common-rank Monte Carlo and paired physical-configuration cluster-bootstrap outputs; fold-aligned permutation-importance results; compact result-source tables; documentation; pinned dependencies; and SHA-256 integrity metadata. The included scripts reproduce the packaged numerical workflow from the supplied design matrix and curated evidence tables. The outputs are conditional screening indicators within the documented partial boundary: the operational carbon burden covers grid electricity and direct diesel/natural-gas combustion, material credits are spend-based proxies, and pack-basis values are cross-model mass-intensity sensitivities. They are not complete plant-level life-cycle assessment or techno-economic assessment results.

Steps to reproduce

1. Download and extract Target_Dependent_EV_LIB_Recycling_Data_and_Code.zip. Open a terminal and change to the extracted JCP_LIB_Recycling_Data_v2 directory. 2. Use Python 3.12; the packaged workflow was verified with Python 3.12.13. On macOS or Linux, create and activate an isolated environment: python3 -m venv .venv source .venv/bin/activate On Windows, use Windows Subsystem for Linux (WSL) or another Bash environment. The supplied full-workflow launcher is a Bash script and is not intended for Command Prompt or PowerShell. 3. Install the pinned dependencies: python3 -m pip install -r requirements.txt If XGBoost or LightGBM reports a missing OpenMP runtime, install the appropriate system runtime and retry; for example, use brew install libomp on macOS. 4. From the package root, run the complete tabular workflow: bash code/run_reproducible_pipeline.sh 5. By default, the script creates reproduced_full_workflow beside the extracted package. It does not modify the packaged result tables and refuses to overwrite an existing destination. To use another new destination, run, for example: bash code/run_reproducible_pipeline.sh ../reproduced_full_workflow_2 6. The strict verifier runs automatically as the final stage. A successful run reports "status": "PASS" and then prints: Reproduced workflow outputs: .../data/processed 7. The workflow starts from the supplied scenario design matrix and curated evidence tables. It reproduces the packaged numerical workflow, including deterministic targets, cell- and pack-basis model evaluation, uncertainty analyses, interpretability outputs, sensitivity diagnostics, and grouped adjusted-balance validation. It does not reproduce source-document extraction or article typesetting.

Institutions

• Wuhan University of Science and Technology

  Hubei, Wuhan

Categories

Licence