Real-Time Stochastic Load Monitoring Dataset

The dataset was collected at the Smart Grid and Power Lab, US Pakistan Center for Advanced Studies in Energy, National University of Sciences and Technology using a PMU/PDC-based real-time monitoring system integrated with NI LabVIEW for synchronized electrical measurements. The primary objective of this dataset is to support research in Non-Intrusive Load Monitoring (NILM), smart grid analytics, and machine learning-based appliance classification.
Research Hypothesis

The central hypothesis of this dataset is that distinct electrical appliances exhibit unique and learnable power consumption signatures that can be accurately identified using time-series machine learning and deep learning models, even in real-time and noisy smart grid environments. Furthermore, it is assumed that PMU-synchronized measurements (voltage, current, and real power) improve the separability of appliance load patterns compared to conventional unsynchronized smart meter data.

What the Data Shows

The dataset contains time-stamped electrical measurements including voltage, current, real power, and aggregated power consumption for multiple residential and laboratory appliances such as TV, refrigerator, microwave oven, vacuum cleaner, 6-ton AC, iron, kettle, toaster, washing machine, coffee machine, hair dryer, and desktop computer.

Key characteristics observed in the data include:

Strong variability in power consumption across different appliances
Repetitive and stable load signatures for high-consumption devices (e.g., refrigerator, AC)
Highly transient behavior for resistive loads (e.g., kettle, iron, toaster)
Mixed operational patterns due to stochastic switching and real usage conditions
Clear temporal structure suitable for sequential modeling
Each record in the dataset represents a real-time measurement of electrical parameters associated with a specific appliance operating state. The dataset can be interpreted as a multi-class time-series classification problem where:

Input features: voltage, current, real power, aggregated power
Target label: appliance type

The data can also be reformulated for:

Sequence prediction (load forecasting)
Anomaly detection in power usage
Feature extraction for NILM disaggregation
Hybrid physics-ML modeling of energy consumption