遇见数据集

REPRODUCIBILITY PACKAGE When Do World Models Help Energy-Aware Cooling at Cellular Base Stations? Interventional Forecasting and Physics-Guided Closed-Loop Planning

收藏
Zenodo2026-11-01 更新2026-11-05 收录
官方服务:

资源简介:

This repository provides supporting research materials for “When Do World Models Help Energy-Aware Cooling at Cellular Base Stations? Interventional Forecasting and Physics-Guided Closed-Loop Planning.” The research investigates whether action-conditioned world models predict temperature and power accurately under alternative cooling plans, and whether these forecasts improve closed-loop decisions at cellular base stations. It compares latent and physics-guided world models with forecasting baselines, operator rules and adaptive resistance–capacitance models, including models with compressor-protection logic. The deposited materials include simulation code, model implementations, experiment configurations, derived results and analysis scripts for reproducing the reported tables, figures and statistical comparisons. The evaluation covers interventional forecasting, planning under distribution shift, adaptation to new sites, and trade-offs between energy consumption, emissions and thermal exposure. The findings show that improved forecasting does not necessarily produce better control. Physics-guided world models outperform several baselines when important mechanisms are missing from the physics model, while protection-aware adaptive physics provides the strongest planning performance in most tested settings. Interventional ground truth and closed-loop control results are simulation-based. Public base-station energy and server-room datasets provide complementary factual forecasting evidence; proprietary operator records support steady-state and benchmark-assumption checks. No operational cooling savings from a deployed field trial are claimed. Original public datasets remain available from their respective sources and are subject to their original licences. Proprietary operator-level records are not redistributed. Consult the README for package contents, dependencies, execution instructions and reproduction procedures.

提供机构:
Zenodo
创建时间:
2026-11-01
二维码
社区交流群
二维码
科研交流群
商业服务