NILM Data-Set for Varying Operating Voltages

Citation Author(s):
Submitted by:
Raghunath Reddy
Last updated:
Fri, 08/07/2020 - 03:00
0 ratings - Please login to submit your rating.


Globally building sector energy consumption is increasing rapidly. Improving building energy efficiency is essential for sustainability. Load monitoring provides detailed consumption feedback to enable consumers to save energy. Non-Intrusive Load Monitoring (NILM) is a cost-effective way to identify individual appliance energy consumption from aggregate energy consumption. Machine learning-based NILM techniques have been proposed in the literature to accurately identify appliances.In the real-world, the operating conditions may vary from training conditions affecting the performance of supervised NILM techniques. The performance of supervised NILM techniques degrades in presence of data variations due to noise, sensor drift, or source voltage fluctuations. This study evaluates the performance of supervised event detection and appliance identification techniques under varying operating voltages. An automated setup captures the aggregate consumption data of various home appliances. We employ steady state-supervised event detection and appliance identification using standard learning algorithms such as K-NN, Naive Bayes, Decision Tree, and Random Forest.The results show that varying voltage data significantly affects the performance of classifiers. We also evaluate mitigation strategies such as normalization, feature selection, and class balancing for improving classifiers. Insights from the experimental results help in developing robust NILM systems that can overcome the effects of data variations.


There are 6 files corresponding to aggregate energy data collected at 6 different source voltages from 190V to 240V.The features in the data are Voltage,Current, Active Power(P), Apparent Power(S), Reactive Power(Q), Power Factor, Phase angle, Class, Resistance, Pdiff, Sdiff, Qdiff, Rdiff, Device.It is a continuous time series data collected at 10 Hz from automated data collection setup. Collect aggregate data of 7 home appliances namely Geyser, Kettle, Mixer, Oven, Fan, Air Purifier and Vacuum Cleaner. The Class column contains a numeric value whose binary equivalent represents the configuration of appliance in ON/OFF state that results the aggregate data. The numbers are grey code sequence, allowing only one device change state at any time. Device column contains the label of the event happened in data such as Geyser in ON state event labelled as GON. Fan in OFF state event is FOFF. The No events data are labelled as NN. Resistance column is calculated feature, is the ratio of voltage and current. Other features such as Pdiff, Sdiff, Qdiff and Rdiff are the single difference values of P, S,Q and R features. 



Submitted by Nevena Musikic on Fri, 07/23/2021 - 01:22

Dataset Files

Open Access dataset files are accessible to all logged in  users. Don't have a login?  Create a free IEEE account.  IEEE Membership is not required.