Data and results for "Reservoir Conformal Prediction for Greenhouse Climate Forecasting under Seasonal and Control-Induced Distribution Shifts"

Processed datasets, trained forecasters, prediction intervals and closed-loop simulation traces for the manuscript "Reservoir Conformal Prediction for Greenhouse Climate Forecasting under Seasonal and Control-Induced Distribution Shifts" (submitted to IEEE Access). Code: https://github.com/lsngmin/rescp-greenhouse (https://doi.org/10.5281/zenodo.23754280). Unzip each archive in the root of the code repository; the paths below are relative to that root. README.txt repeats this description. data.zip (1 MB)- data/kangwon.parquet, data/agc1_ref.parquet: 10-min tables with nine inputs (indoor T/RH, outdoor T/RH/wind, irradiance, vent opening, hour sin/cos), four next-hour targets, a valid flag and the split label- data/splits.json: train / calibration / test dates results.zip (152 MB)- results/<dataset>/<model>/seed<k>/: pred.csv (point forecasts for calibration and test), model files, metrics.json, rescp_{T,RH}_a90_w2000_T0.3.csv (ResCP intervals), rescp_metrics.json, base_{swcp,aci}_{T,RH}.csv (sliding-window CP and ACI intervals), base_metrics.json- results/<dataset>/<model>/: tuning.json, rescp_summary_a90_w2000_T0.3.json, base_summary.json results_sim.zip (106 MB)- results_sim/baseline_2011.csv: open-loop run of the rule-based controller in GreenLight-Gym, Amsterdam 2011- data_sim/sim_2011.parquet, splits_sim_2011.json: simulation dataset in the same format as the measured data- results_sim/pred/sim_2011/<model>/: forecasters trained on the simulation dataset- results_sim/closed_2011/<mode>/<model>/seed<k>/{trace.csv, metrics.json}: closed-loop runs; mode = baseline, D2 (interval-based control), P / P2 / P3 (point-forecast control with cold margin 0 / 1.0 / 1.5 °C)- results_sim/closed_2011/{all_runs, summary}.csv figs.zip (2 MB)- figs/: figures of the paper, PDF and PNG Datasets: kangwon = Korea, sweet pepper greenhouse, 2023 (Mendeley Data doi:10.17632/mwg293k2r9.2, CC BY 4.0; outdoor variables from KMA ASOS station 211, KOGL type 1). agc1_ref = Netherlands, Autonomous Greenhouse Challenge 2018, Reference compartment (4TU.ResearchData doi:10.4121/uuid:e4987a7b-04dd-4c89-9b18-883aad30ba9a, CC0).Forecasters: lgbm, gru, xlstm, moderntcn, timemixer, patchtst, itransformer; seeds 0–4.