SafeLeak-RCD: Residential Residual Current Decomposition Benchmark

This repository contains SafeLeak-RCD, the public benchmark bundle prepared for the manuscript:

Physics-Regularized Conditional Flow Matching for Branch-Conditioned Residual Current Decomposition in Electrical Safety Monitoring

Release Contents

• benchmark/ The exact train/validation/test split used in the manuscript revision.
• processed_entities/ The processed per-entity bundle used to construct the benchmark and inspect the augmentation pipeline.

Task

Given:

• aggregate residual current
• aggregate active power
• a target-branch power cue

predict the residual current of the selected branch.

Benchmark Summary

• single-phase residential electrical safety monitoring
• 12 branches
• 1-minute target interval
• 7 entity-level panels in total
• entity-disjoint train/validation/test = 5/1/1
• train split: 104,835 rows total with synthetic variants restricted to training only
• validation split: 7,091 real rows
• test split: 11,991 real rows

Split Files

• benchmark/train.csv
• benchmark/validation.csv
• benchmark/test.csv
• benchmark/split_config.json

Example Row

Each CSV row is one timestamped panel snapshot. The benchmark stores the aggregate channels and all branch channels in the same row. A selected real example from benchmark/train.csv looks like this:

{
  "timestamp": "2024-08-01 00:10:00",
  "total_residual_current": "13.32345",
  "total_power": "20632.769769",
  "branch_1_power": "2210.233346",
  "branch_1_current": "0.0036",
  "branch_2_power": "190.9009",
  "branch_2_current": "0.1582",
  "branch_3_power": "0.0",
  "branch_3_current": "0.0003",
  "branch_12_power": "3974.763915",
  "branch_12_current": "2.774",
  "synthetic_variant": 0,
  "segment_id": "shanse001_aug_chunk_01_1min_base",
  "entity_id": "shanse001_aug_chunk_01_1min"
}

Branches 4 to 11 are omitted above for brevity. For branch-conditioned learning, pick one branch index k and map:

target branch: 12
inputs  = (total_residual_current=13.32345, total_power=20632.769769, branch_12_power=3974.763915)
target  = branch_12_current=2.774

Processed-Entity Bundle

• processed_entities/manifest.json lists every released entity bundle.
• Each entity directory contains:
  • *.base.csv for the real observed base panel
  • *.variant_*.csv for synthetic augmentation variants
  • *.csv for the combined training-time view used by the benchmark builders
  • *.metadata.json for dataset-level statistics and notes
• All paths stored in manifest.json and *.metadata.json are repo-relative paths inside this dataset repository.

Loading Example

from datasets import load_dataset

dataset = load_dataset("haayan/safeleak-rcd", "benchmark_split")
print(dataset["train"])

Access and Use

This release is intended to support research reproducibility for the associated manuscript. Check the license field and any additional usage notice configured on the Hugging Face repository before downstream redistribution or commercial use.

License

This dataset is released under CC-BY-NC-4.0. This means downstream users may:

• download and reuse the data
• redistribute derived copies
• adapt the benchmark for follow-up work
• use it for non-commercial research and educational purposes with attribution

Commercial use is not permitted without separate permission from the dataset authors.

Citation

If you use this dataset, cite both:

• the manuscript Physics-Regularized Conditional Flow Matching for Branch-Conditioned Residual Current Decomposition in Electrical Safety Monitoring
• the dataset repository https://huggingface.co/datasets/haayan/safeleak-rcd