haayan/safeleak-rcd

SafeLeak-RCD是一个用于住宅剩余电流分解的基准数据集，旨在支持电气安全监测研究。该数据集基于单相住宅电气安全监控系统，包含12个分支，时间间隔为1分钟，总共有7个实体级面板。数据集的任务是：给定总剩余电流、总有功功率和目标分支功率提示，预测选定分支的剩余电流。数据集分为训练、验证和测试集，其中训练集包含104,835行数据（包括仅限于训练的合成变体），验证集包含7,091行真实数据，测试集包含11,991行真实数据。数据以CSV文件格式提供，每行代表一个时间戳的面板快照，包括总通道和所有分支通道。数据集还包括处理后的实体包，用于构建基准和检查增强管道。该数据集适用于表格和时间序列分析，支持非侵入式负载监测（NILM）和物理信息学习等任务。SafeLeak-RCD is a benchmark dataset for residential residual current decomposition, designed to support research in electrical safety monitoring. It focuses on single-phase residential electrical safety monitoring, with 12 branches, a 1-minute target interval, and a total of 7 entity-level panels. The task is: given aggregate residual current, aggregate active power, and a target-branch power cue, predict the residual current of the selected branch. The dataset is split into train, validation, and test sets, with the train split containing 104,835 rows total (including synthetic variants restricted to training only), the validation split containing 7,091 real rows, and the test split containing 11,991 real rows. Data is provided in CSV format, where each row represents a timestamped panel snapshot with aggregate channels and all branch channels. It also includes a processed-entity bundle for benchmark construction and augmentation pipeline inspection. The dataset is suitable for tabular and time-series analysis, supporting tasks such as non-intrusive load monitoring (NILM) and physics-informed learning.