AC Power Consumption Dataset

Citation Author(s):
Jie
Su
Zhen
Hong
Lei
Ye
Tao
Liu
Sizhuang
Liang
Shouling
Ji
Gagangeet
Aujla
Reheem
Beyah
Zhenyu
Wen
Submitted by:
Jie Su
Last updated:
Wed, 11/22/2023 - 21:33
DOI:
10.21227/v2vf-3r56
License:
389 Views
Categories:
Keywords:
0
0 ratings - Please login to submit your rating.

Abstract 

<p>This dataset offers a unique compilation of AC power consumption signals from four different computing devices, encompassing a variety of software applications. Each entry in the dataset meticulously records the intricate fluctuations in power usage as specific software programs are initiated and terminated, providing a 40-second window into the energy profile of each application. With a high-resolution sampling rate of 44100Hz, the dataset captures the nuanced energy demands placed on the electrical infrastructure, conforming to the 220V, 50Hz standard. Tailored for researchers delving into the realm of non-intrusive profiling and power side-channel analysis, this dataset serves as a foundational tool for advancing the detection of software activity through power consumption patterns. The diversity of devices and operating systems included—ranging from various Windows laptops to Linux desktops—presents a comprehensive resource for developing and benchmarking algorithms aimed at identifying and categorizing power usage signatures within the growing field of cyber-physical system security.</p>

Instructions: 

#######################################################################

Power Consumption Dataset 1.0

2020-5-6

#######################################################################

 

Introduction:

This dataset contains AC power consumption data of computers running various software and is intended solely for scientific research. The dataset includes four devices — ThinkPad (Windows), Lenovo (Windows), ASUS (Windows), Desktop (Linux). The number of software applications collected on each device varies. Under each device, the data is categorized by the name of the software. 

The data is in audio format, with a recording duration of 40s. The software starts at the 5th second and closes at the 35th second, with 5 seconds before and after left out to allow the computer to return to its normal operational state.

The data sampling frequency is 44100Hz. The AC power involved in signal collection is 220V, 50Hz (China's standard AC power).

 

#######################################################################

 

Folder Descriptions Related to Data:

eff_data_ThinkPad: Power consumption data related to ThinkPad (Windows)

eff_data_lenovo: Power consumption data related to Lenovo (Windows)

eff_data_ASUS1\2: Power consumption data related to ASUS (Windows), split into two parts due to large size; can be merged after decompression

eff_data_linux1\2: Power consumption data related to Desktop (Linux), split into two parts due to large size; can be merged after decompression

continous_test_ThinkPad: Continuous multiple software start-up test data corresponding to ThinkPad (Windows), where the experimenter randomly starts multiple software applications at the rate of one per minute and records for 20 minutes (no corresponding labels, for testing only)

continous_test_lenovo: Continuous multiple software start-up test data corresponding to Lenovo (Windows), similar to the ThinkPad process described above

 

#######################################################################

The following are the process result files used in the current stage of research:

***training_picture: Includes test images of power consumption data for all devices, where audio data is taken from the 5th second onward, cut to a fixed length (5s, 10s, 15s), the sampling rate is compressed to 100Hz (compression method is to take the maximum and minimum values of each power cycle and save them as absolute values), and saved as 224*224 black and white images. The naming format is [original name of the software signal.png].

***testing_picture: Randomly selected 20 pieces of power consumption data for each software, filtered for signal fluctuation starting points (filtering method: calculate the rolling mean of each signal after compression sampling rate, using rolling_mean(100, minwidth=10) to get the signal's moving average curve, then take the overall average value as the threshold, record points above the threshold where the curve's slope is greater than 0 as the starting points for signal fluctuation), and centered on that point, within a range of 0.5s before and after, set a signal start point every 0.1s, cut from the original signal accordingly, compress the sampling rate, and save as 224*224 black and white images. The naming format is [signal start point.png] — which is the estimated time of software start.

***testing_picture_with_noise: Randomly selected 20 pieces of power consumption data for each software, similar to testing_picture, after cutting operations, added specified signal-to-noise ratio noise to each cut signal. The noise is irregular Gaussian white noise. Compress the sampling rate, save as 224*224 black and white images. The naming format is [signal start point.png] — which is the estimated time of software start.