🏠 WHITED – Worldwide Household and Industry Transient Energy Dataset

The WHITED dataset contains high-frequency voltage and current recordings of household and small-industry appliances from multiple regions worldwide.
Each sample captures the start-up transient of an appliance, enabling research in non-intrusive load monitoring (NILM), appliance classification, and regional grid signal analysis.

---

📘 Dataset Summary

• Appliances: 110 devices
• Appliance types: 47 classes
• Regions: 6 (4 in Germany, 1 in Austria, 2 in Indonesia)
• Sampling rate: 44.1 kHz
• Bit depth: 16-bit
• Duration per recording: 5 s
• Channels: 2 (current, voltage)
• File format: .flac (lossless audio)

Each file records an appliance start-up transient and is named as follows:

[Class]_[Name]_[Region]_[#Kit]_[Timestamp].flac
Example: GuitarAmp_Marshall8240_R3_MK2_20151115133402.flac

Region:

• R1: Germany, Garching
• R2: Germany, Munich
• R3: Germany, Gotha
• R4: Indonesia, Bandung, Jalan Raya Sapan
• R5: Indonesia, Bandung, Jalan Palem Botol
• R6: Austria, Grieskirchen
• R7: Germany, Gohrisch
• R8: Canada, Burnaby

#Kit	voltage factor	current factor
MK1	1033.64	61.4835
MK2	861.15	60.200
MK3	988.926	60.9562

---

⚙️ Measurement Setup

A low-cost, portable sound card-based measurement system was used for data collection, consisting of:

• Modified 3-port extension cord
• YHDC current clamp (1 V @ 30 A primary current)
• AC-AC transformer + voltage divider (230 V → 0.47 V)
• External USB sound card (C-Media CM6206 chipset)

Each setup was individually calibrated using a VOLTCRAFT VC-330 multimeter.
Three identical measurement kits were built for recordings in different locations.

---

🗂 Data Structure

./
├── Experiments/
│   ├── 2Appl/
│   ├── ACAC
│   ├── Compressing
│   ├── in_household_trained_isolated
│   ├── readme.txt
│   └── ...
│
├── images/
│   ├── Kettle_Bosch_label.jpg
│   └── VacuumCleaner_Philips_label.jpg
│
├── MIXED/
│   ├── Heater_Multitool_MK1_20151202171103.flac
│   ├── Heater_Multitool_MK1_20151202171450.flac
│   └── ...
│
├── notUsed/
│   ├── LEDLight_RunningTestTrack_20151102061840.flac
│   ├── LEDLight_RunningTestTrack_20151102061851.flac
│   └── ...
│
├── plots/
│   ├── AC_Electrolux_powerTimeDomain.png
│   ├── BenchGrinder_KinFujiKBG2080_powerTimeDomain.png
│   └── ...
│   
├── AC_Electrolux_r5_MK2_20151031065948.flac
├── AirPump_SimexHB-135C_r2_MK1_20160124164919.flac
├── BenchGrinder_KinFujiKBG2080_r4_MK2_20151102052641.flac
├── CableModem_MotorolaSB5102_r8_MK1_20160515233311.flac
├── CFL_IKEA15W_r1_MK2_20151012103604.flac
├── ...

Each .flac file stores a stereo recording:

• Left channel: current
• Right channel: voltage

---

🔬 Research Applications

WHITED supports tasks in:

• Non-Intrusive Load Monitoring (NILM)
• Appliance classification
• Transient signal analysis
• Grid region fingerprinting
• Machine learning on waveform data

Reported benchmark results:

• Appliance classification: 95 % accuracy (SVM, 10-fold CV)
• Region classification: 99 % accuracy (voltage-based features)

---

📊 Technical Specifications

Parameter	Value
Sampling rate	44.1 kHz
Bit depth	16-bit
Effective SNR	75.9 dB
Current resolution	13.5 mA
Voltage step	0.313 V
Power step	3.1 W
Duration per sample	5 s
Format	FLAC (stereo)

Appliances with a minimum power consumption of 20 W are included, covering most common household and light industrial devices.

---

🧺 Appliance Classes (47)

Appliance Type	Count of Appliances
AC	1
Air Pump	1
Bench Grinder	1
CFL	2
Charger	7
Coffee Machine	1
Deep Fryer	1
Desktop PC	1
Desoldering tool	1
Drilling Machine	2
Fan	6
Fan Heater	1
Flat Iron	2
Game Console	4
Guitar Amp	1
Hair Dryer	6
Halogen Fluter	1
Heater	1
HiFi Rack	1
Iron	3
Jigsaw	1
JuiceMaker	1
Kettle	6
Laptop	1
Laserprinter	1
LED Light	9
Light bulb	6
Massage tool	3
Microwave	2
Mixer	4
Monitor	2
Mosquito Repellent	1
Multitool	1
Powersupply	4
Projector	1
Sewing Machine	1
Shoe warmer	2
Shredder	2
Soldering Iron	2
Toaster	4
Treadmill	1
TV	1
Vacuum Cleaner	4
Washing Machine	1
Water Heater	4
Water Pump	1

---

🌍 Regional Distribution

• Germany: 4 regions
• Austria: 1 region
• Indonesia: 2 regions
• Canada: 1 region

Each region’s voltage waveform reflects its distinct grid characteristics.

---

🧪 Example Use

from datasets import load_dataset
import soundfile as sf

ds = load_dataset("MatthiasKahl/whited", split="train")
audio, sr = sf.read(ds[0]["audio"]["path"])

print(ds[0]["file_name"])
print(f"Sample rate: {sr}, Shape: {audio.shape}")

---

🔬 Experiments

In household trained (isolated)

• typical household devices
• only one device per class
• isolated measurment (without any other appliance)
• x-fold crossvalidation
• WHITED, PLAID

In household trained (isolated, sparse)

• typical household devices
• only one device per class
• only 2 training measurements per class
• rest as testing
• WHITED, PLAID

In household trained (real world)

• typical household devices
• only one device per class
• real measurment from UK-Dale
• x-fold crossvalidation
• UKDALE, REDD

Generally trained (isolated)

• typical devices
• multiple devices per class
• isolated measurements from WHITED
• leave one appliance out evaluation
• WHITED, PLAID

---

📄 Citation

If you use this dataset, please cite:

Matthias Kahl, Anwar Ul Haq, Thomas Kriechbaumer, and Hans-Arno Jacobsen.
WHITED – A Worldwide Household and Industry Transient Energy Dataset.
IEEE Conference on Data Mining Applications in Sustainability (SustKDD), 2016.
DOI: 10.13140/RG.2.2.19590.11844

---

🧑‍🔬 Authors

• Matthias Kahl – Technical University of Munich
• Anwar Ul Haq – Technical University of Munich
• Thomas Kriechbaumer – Technical University of Munich
• Hans-Arno Jacobsen – University of Toronto

---

🪪 License

The dataset is released for research and educational purposes under the Creative Commons Attribution 4.0 International (CC BY 4.0) license.
You must give appropriate credit to the original authors when using or redistributing the data.

---

📥 Access

Paper: WHITED – A Worldwide Household and Industry Transient Energy Dataset (2016)