Real-Time Stochastic Load Monitoring Dataset
Description
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
Files
Steps to reproduce
The dataset was generated using a controlled yet realistic experimental setup where appliances were individually operated under monitored conditions. Data acquisition followed these steps: PMU-based synchronized measurement collection Real-time streaming via PDC system LabVIEW-based acquisition and logging CSV export for preprocessing and machine learning workflows Timestamp alignment and basic cleaning for consistency
Institutions
- National University of Sciences and TechnologyIslamabad, Islamabad