Lights in the Night
Outage Identification using Remote Sensing in India
L. Vreedenburgh (TU Delft - Technology, Policy and Management)
Nihit Goyal – Mentor (TU Delft - Organisation & Governance)
T. Verma – Graduation committee member (TU Delft - Policy Analysis)
M.E. Warnier – Coach (TU Delft - Multi Actor Systems)
More Info
expand_more
Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.
Abstract
Electricity is essential in the modern world. Although India is near reaching 100% electrification, for many of its citizens, reliable access to electricity is still a major issue. It is not clear how, when, and where the reliability issues arise. Previous studies have not shed light on electricity outages at a granular level. Furthermore, identifying outages is difficult without proper metering.
This research aims to acquire insight into how electrical outages vary during the nighttime in the Uttar Pradesh, one of the states experiencing the most outages in India. Remote sensing data in the form of nighttime radiance, wind, precipitation, temperature, air quality, land cover, and population are used to identify outages, as they have been identified in the literature to be influencing or correlated with outages. Using a Random Forest (RF) classification algorithm, outages are identified in India for the year 2018. The RF is trained on Electricity Supply Monitoring Initiative (ESMI) real-time electricity household senors. RF is cross-validated using multiple strategies to accurate measure the performance and to ensure no data leakage. Using the resampling technique Synthetic Minority Oversampling technique (SMOTE), the performance of the RF is increased for outage classification. RF is used to classify three classes: Never Access, Normal Access, and Outage. The distribution of these three classes is highly imbalanced, with the Outage class being in the minority.
To further validate the methodology, the spatial and temporal sampling was done during cross-validation. Using this method, 91% of Never Access samples were categorized as such, with a precision of 84%. 69% of Normal Access samples were categorized as such, but with 84% precision. The performance of the outage classification is the worst. Nonetheless, 55% of the Outage samples were categorized as such, although with a precision of just 39%. A map is created for Uttar Pradesh showing the results of the rates for each of the classes in 2018 between 00:00 and 02:00 with a spatial granularity of 0.1° x 0.1°. The map indicate that the electricity usage during the nighttime in Uttar Pradesh is in large parts of the state non-existent. The map be used as a precursor for future field work and help policymaker and researchers identify reliability and fairness issues.