PowerZooJax/PowerZooDataset

PowerZoo数据集包含真实世界的电力系统和数据中心时间序列数据、标准电网拓扑结构以及连接两者的JSON清单，专为强化学习和预测研究设计，与PowerZoo/PowerZooJax基准代码（单独发布）配合使用。具体内容包括：1. 11个Parquet时间序列文件，来源于公共监管机构和云服务提供商发布的数据（涵盖英国/澳大利亚的电力负荷、燃料类型发电量、日前预测、市场中间价格；阿里巴巴/Azure/Google数据中心利用率）。2. 14个电网案例文件（Python类，包含总线/分支/发电机/负荷表），覆盖传输系统（5至2383个总线）和配电系统（33至533个总线）。3. 11个JSON清单，将每个Parquet文件的原始列映射到共享标准模式（例如，OPERATIONAL_DEMAND → load.actual_mw），以便不同来源的轨迹在实验中清晰组合。数据集适用于电力系统、智能电网、时间序列预测、强化学习等任务。The PowerZoo dataset contains real-world power system and data center time series data, standard power grid topologies, and JSON manifests that link the two. It is specifically designed for reinforcement learning and forecasting research, and is compatible with the separately released PowerZoo/PowerZooJax benchmark code. The detailed components are as follows: 1. Eleven Parquet time series files, sourced from datasets published by public regulatory authorities and cloud service providers. These files cover power load, fuel-type power generation, day-ahead forecasts, and market mid prices in the UK and Australia, as well as data center utilization rates of Alibaba, Azure, and Google. 2. Fourteen power grid case files (Python classes containing bus, branch, generator, and load tables), covering transmission systems (ranging from 5 to 2383 buses) and distribution systems (ranging from 33 to 533 buses). 3. Eleven JSON manifests that map the original columns of each Parquet file to a unified standard schema (e.g., OPERATIONAL_DEMAND → load.actual_mw), allowing clear combination of trajectories from diverse sources in experimental settings. This dataset is suitable for tasks such as power systems, smart grids, time series forecasting, and reinforcement learning.