Understanding long-term energy use in off-grid solar home systems in sub-Saharan Africa

Solar home systems provide low-cost electricity access for rural off-grid communities. As access to them increases, more long-term data becomes available on how these systems are used throughout their lifetime. This work analyses a dataset of 1,000 systems across sub-Saharan Africa. Dynamic time warping clustering was applied to the load demand data from the systems, identifying five distinct archetypal daily load profiles and their occurrence across the dataset. Temporal analysis reveals a general decline in daily energy consumption over time, with 77% of households reducing their usage compared to the start of ownership. On average, there is a 33% decrease in daily consumption by the end of the second year compared to the peak demand, which occurs on the 96th day. Combining the load demand analysis with payment data shows that this decrease in energy consumption is observed even in households that are not experiencing economic hardship, indicating there are reasons beyond financial constraints for decreasing energy use once energy access is obtained.

💡 Research Summary

This paper presents a large‑scale, longitudinal analysis of solar home systems (SHS) deployed in sub‑Saharan Africa, focusing on daily load‑demand patterns, their evolution over time, and the relationship between energy consumption and pay‑as‑you‑go (PAYG) payment behavior. The authors obtained a dataset from the off‑grid energy provider BBOXX comprising 1,000 households equipped with a 50 W PV panel, a 12 V 17 Ah lead‑acid battery, and a set of typical low‑income appliances (lighting, phone chargers, fans, radios, and televisions). The raw measurements were recorded at irregular intervals whenever a change in voltage or current exceeded a preset threshold, with a 10‑minute fallback. These measurements were converted to hourly energy consumption using a zero‑order hold and numerical integration, yielding 647,021 daily load profiles spanning up to 1,230 days of operation. To make the computational problem tractable, the authors performed stratified sampling: first selecting ten representative days per household (10,000 series) and then further reducing the set to 2,000 series while preserving the overall distribution of daily energy use.

For clustering, the study employs a novel DTW‑MIP (Dynamic Time Warping – Mixed‑Integer Programming) approach previously described by Kumtepeli et al. DTW provides an elastic distance metric that aligns time‑series despite temporal shifts, while the MIP formulation guarantees a globally optimal assignment of series to clusters, avoiding the local‑optimum traps common to k‑medoids or k‑means. No warping window is imposed, allowing the algorithm to capture long‑range similarities such as whether peaks occur in the morning or evening. The optimal solution identified five archetypal daily load profiles, each representing a distinct usage pattern: (1) minimal load dominated by lighting and phone charging, (2) fan and radio use with an afternoon peak, (3) evening‑centered TV + fan consumption, (4) higher‑intensity loads including refrigeration or pumps, and (5) very low or near‑zero consumption days. After determining the centroids, every daily profile in the full dataset was assigned to its nearest cluster, producing a chronological sequence of cluster labels for each household.

Temporal analysis of these sequences revealed a consistent decline in daily energy consumption after an initial peak occurring around day 96. On average, consumption fell by 33 % by the end of the second year relative to that peak. Moreover, 77 % of households reduced their usage compared with the start of ownership. To investigate whether financial constraints drove this reduction, the authors linked the load data with payment records. BBOXX’s PAYG model provides a “remaining credit” time series, which the authors converted to “days of electricity remaining” to standardize across currencies. They defined an “utilisation rate” as the proportion of days a household was not in an economic outage (i.e., not lacking credit). Even households with high utilisation rates—indicating few or no economic outages—exhibited the same downward consumption trend. This suggests that factors beyond pure affordability, such as behavioral adaptation, appliance usage optimization, seasonal changes, or evolving perceptions of electricity value, influence long‑term demand.

The paper contributes several important insights. First, it demonstrates that large‑scale, high‑frequency SHS data can be effectively clustered using DTW‑MIP, yielding interpretable load archetypes that can inform system sizing (battery and inverter capacity) and tariff design. Second, the observed post‑adoption consumption decline challenges the common assumption that electrification inevitably leads to ever‑increasing demand; instead, demand may plateau or even recede after an initial adjustment period. Third, the integration of payment data shows that PAYG metrics can serve as early warning signals for potential service interruptions while also revealing that economic hardship is not the sole driver of usage patterns. Finally, by making the full dataset publicly available, the authors provide a valuable resource for the research community to explore further questions about rural electrification, technology adoption, and energy policy.

Limitations include the absence of explicit battery health metrics, climate variables (e.g., solar irradiance, temperature), and detailed household socioeconomic characteristics, all of which could modulate consumption trends. The analysis is also confined to BBOXX customers, which may limit generalizability to other providers or regions. Future work could incorporate multivariate regression or machine‑learning models that jointly consider these additional factors, explore seasonal effects, and assess the impact of targeted interventions such as energy‑efficient appliance distribution or user education programs. Overall, the study offers a robust methodological framework and empirical evidence that can guide both policymakers and off‑grid energy entrepreneurs toward more sustainable and user‑centred SHS deployments.