How much electricity does a typical Sri Lankan home use, and what explains that usage? To answer this properly, we need to look at who lives in the house, which appliances they run, how big the home is, and, if possible, the materials that have been used to build it. Understanding what drives their electricity bill up will help families spot energy waste and lower their monthly bills. On a larger scale, it helps energy planners roll out the right energy-saving technologies nationwide, while giving policymakers and regulators the data they need to set fair electricity rates and direct subsidies to the people who need them most.
However, collecting this data at the scale and level of detail needed is difficult, and existing datasets reflect the practical trade-offs involved. Researchers have to choose between detailed contextual and smart-meter data from a few homes and fewer data points from a larger sample. Until recently, no household electricity consumption dataset was widely available for Sri Lanka, even with these limitations. Released by LIRNEasia in 2025 with support from the Lacuna Fund and in partnership with Lanka Electricity Company (LECO), the Residential Electricity Consumption Dataset for Sri Lanka (RECON-SL) bridges this gap. It is the first dataset of its kind in the country—and one of the few globally—to pair electricity records with repeated surveys across more than 4,000 homes. The dataset is publicly available and designed to support energy policymaking, social research, and machine learning applications. (Scroll down to access the dataset.)
Championing new data for better policy
Good policy depends on good data. Policymakers rely on evidence to understand the world around them, spot problems, decide how to respond, and measure whether their solutions actually work. National statistical offices (NSOs) and government agencies produce much of this foundational data, but their resources are limited. Even the best-funded NSOs cannot keep pace with every emerging question or explore every new source of data.
LIRNEasia has long helped bridge this gap by pairing traditional statistics with alternative data sources to guide public policy. For instance, we have analyzed mobile call detail records (CDRs) to inform urban transport planning and public health responses, and applied machine learning to satellite imagery to map crop cultivation and help target social assistance for poverty alleviation.
Beyond answering immediate policy questions, we also build foundational datasets for the broader AI and research community. Our teams have developed linguistic corpora to advance local-language AI, alongside curated datasets—such as our annotated Sinhala misinformation corpus—to support research at the intersection of AI and information integrity. As modern analytical tools unlock new ways to learn from data, access to high-quality, representative datasets becomes even more critical. Open datasets give researchers, policymakers, and developers a shared foundation to explore urgent questions and test new ideas. RECON-SL is part of LIRNEasia’s broader effort to make new sources of data available for research and public policy.
A bird’s-eye view of RECON-SL
RECON-SL has two main parts: electricity consumption records and household surveys conducted over time.
Figure 1: Structure of the RECON-SL dataset
1. Electricity consumption records
The dataset contains electricity consumption records for 4,063 households, collected between October 2022 and December 2024.
All participating households have monthly consumption records, collected at the end of each billing cycle. These provide a view of how electricity usage changes from month to month.
Of these households, 1,438 have smart meters that record electricity use every 15 minutes. These high-frequency readings capture changes in electricity use throughout the day that monthly records are unable to reflect.
For the smart-metered households, the dataset also includes readings at six-hour intervals. These came from a separate dataset maintained by LECO and help supplement gaps caused by a storage system error that affected some of the 15-minute records.
2. Longitudinal survey data
To understand the households behind the meter readings, we conducted three rounds of surveys, known as Wave 1, Wave 2, and Wave 3, between December 2023 and January 2025.
Wave 1 established the baseline data. It covered all 4,063 households and collected information about residents, incomes and other socioeconomic characteristics, appliances and their use, rooms, and household energy systems.
Wave 2 tracked the changes since the last visit. We revisited the same households to capture the changes in appliances, rooms, residents, and electricity-use habits.
Wave 3 added questions about the energy usage experiences. Alongside tracking further changes in the houses, we asked about electricity affordability, views on tariffs, and energy security.
Across its components, RECON-SL contains more than 50-million-meter readings and responses from over 11,000 household interviews.
Its value comes from connecting the two complementary sources: electricity consumption records with the contextual information on the households. Researchers can examine electricity use at different time intervals alongside information about household members, housing, appliances, and people’s views and experiences.
RECON-SL is the first dataset of its kind in Sri Lanka and one of the few worldwide to link smart meter records with repeated household surveys at this scale. It offers a resource for studying energy use in a South Asian setting, with applications in public policy, energy research, and machine learning.
Figure 2: Sampling and data collection process for RECON-SL
What can household electricity data tell us?
Electricity records tell us how much energy a household uses. Linking those records with household surveys adds more context to the data – the kind of people that live in the house, how they use appliances throughout the day, what the house is like, and how they experience electricity costs.
That combination opens up several areas of research.
- Designing better energy policies.
Matching electricity supply with demand is a challenge, particularly in countries where infrastructure and resources are limited. In South Asia, stark differences in household incomes and living conditions add to that challenge. These differences make it harder to design energy policies that work fairly for everyone.
RECON-SL can help researchers and policymakers explore these differences. They can examine electricity use across households with different incomes and backgrounds, look for seasonal patterns, and study how tariff changes affect different groups.
- Understanding the effects of weather and climate.
Household electricity use can change with the weather. Understanding that relationship can help researchers explore how future climate conditions might affect the future electricity demand.
RECON-SL can support this work by providing consumption data that can be analyzed alongside weather records to explore what changing weather conditions mean for future power needs.
- Developing and testing machine learning models.
Machine learning offers ways to find patterns in large datasets and make predictions. But developing useful models requires high-quality data to train and test them.
RECON-SL can be used to develop and test machine learning models on tasks such as forecasting electricity demand, estimating appliance ownership, and grouping households with similar consumption patterns.
Get started with RECON-SL
RECON-SL is openly available on IEEE DataPort, Kaggle, and Zenodo.
At the time of writing, it had been downloaded 988 times on IEEE DataPort, 375 times on Kaggle, and 83 times on Zenodo. It is encouraging to see people accessing the data, and we look forward to seeing the questions they explore with it.
For a more detailed account of the dataset’s background and contents, the accompanying paper is available on arXiv.
The dataset has also been used in EnergyFM: Pretrained Models for Energy Meter Data Analytics, a peer-reviewed paper on foundation models for analysing energy meter data.
If you work on energy policy, household electricity use, or machine learning, we invite you to explore RECON-SL and see how it could support your research.
Acknowledgements
This work was funded by the Lacuna Fund as part of their awards for climate datasets in health and energy (2022 Climate Awardees).
We gratefully acknowledge Lanka Electricity Company (LECO) for providing access to electricity consumption records and facilitating survey data collection.
By Amanda Ariyaratne (Senior Data Science Researcher, LIRNEasia)
References
Sri Lankan Residential Electricity Consumption Dataset, LIRNEasia, Kaggle
Residential Electricity Consumption Dataset for Sri Lanka (RECON-SL), Zenodo
Residential Electricity Consumption Dataset for Sri Lanka (RECON-SL), arXiv, September 23, 2026
