From June 2021 to July 2022, nLine’s grid-sensors were deployed in 150 villages across five counties in Kenya to measure household-level power reliability and voltage quality. This sensor data was provided to researchers at the University of California Berkeley, Yale University, and the University of Pennsylvania seeking to answer a critical research question:
Do stricter donor contracting conditions on electrification projects in low and middle-income countries lead to better power reliability and quality delivered by those projects?
Kenya’s Last Mile Connectivity Project (LMCP), a $600-million nationwide low-voltage-network extension program, provided the ideal real-world scenario for the research team to shed light on this question. Both the World Bank (WB) and the African Development Bank (AfDB) were each quasi-randomly assigned LMCP sites by the Kenyan government to fund and implement the extension of the distribution network to unconnected households within 600 meters of an existing transformer. The academic researcher team aimed to answer the research question by evaluating the impacts of differing contracting and construction procedures between WB and AfDB.
In an earlier blog post, we shared the motivation for this research study and how we worked with the academic team to design and implement a sensor installation sampling strategy that would help answer several research questions, including:
- Is there a significant difference in power outages or in voltage quality between WB and AfDB funded construction sites?
- Are there power quality and reliability differences between grid-connected households that are located closer or further from the central transformer?
Our previous blog post also summarized engineering lessons learned from the first six months of sensor data collection, including how well our sensors functioned in rural locations with presumed poor cellular connectivity (spoiler: we observed that 90% of data packets from sensors were received within 1 minute after collection and estimated that ~85% of data packets were received on the first transmission attempt, showing a good network connectivity in these remote areas).
In this blog post, we share summary statistics of power quality and reliability sensor data collected from the grid across 150 villages and how this data informed the key research questions described above. View the full research study and published results here.
We wish to thank our generous funders USAID Power Africa and the East Africa Energy Program, implemented by RTI International as well as the UK Foreign, Commonwealth & Development Office (FCDO) funded programmes on Energy and Economic Growth (EEG) and Economic Development and Institutions (EDI), without whose support we would not have had the opportunity to test our technology and measurement techniques in a rural context. Thanks to their contribution, we were able to gather a novel and important power reliability dataset from rural parts of Kenya.
Two local project managers were responsible for installing 100 sensors in 150 villages total across five counties in Kenya: Vihiga, Nandi, Kericho, Kisumu, and Kakamega. In each village (referred to as a “site”), four households connected to the same transformer were each given a sensor for two months. During these two months, the sensor collected data on voltage levels and the presence or absence of power at the household’s power outlet. When at least two sensors located under the same transformer experience a loss of power within a short time window of each other, we log this as a power outage. This data collection method afforded both granular and aggregated analyses of grid reliability in the 150 sampled LMCP villages. Learn more about the specifications of the sensor here and our sensing methodology here.
We analyzed 12 months of sensor data (collected in six rotations of two months each) to determine average households’ voltage quality, duration of power outages, and number of outages at both the household-level and the transformer-level. There is a wide range of information and insights that can be drawn from voltage data (for example, voltage sags and spikes can “lead to equipment damage and reduced system safety”). For this project, we chose to study the overall voltage profiles as well as per-site voltage profiles to show voltage variations over time.
We studied the voltage quality across sites during the deployment period. Figure 2 below shows a graph of the percentage of sensed time when voltage was at different levels.
We also visualized the voltage profiles of individual sites to study the pattern of behavior by hour-of-day and by day-of-week. Figure 3 shows the daily and hourly voltage profiles of a set of four randomly selected sites.
The voltage profiles of four randomly selected sites shown in Figure 3 exhibit similar trends amongst the four sensors connected to the same transformer, but there is wide variation in voltage quality across LMCP villages. We notice some sites consistently experiencing voltage around the 240V nominal voltage (e.g. transformer 1119), some sites mostly experiencing above nominal voltage (e.g. transformer 2235), and other sites regularly measuring below nominal voltage (e.g. transformers 1462 and 100).
For most sites, the voltage profile repeats itself from one day of the week to the next, with no significant difference between weekdays and weekends. This may be explained by the fact that the village-dwellers’ work behavior is the same between weekdays and weekends, as opposed to city-dwellers who would spend most of their weekends in the house hence consuming more electricity over the weekends.
We also observe dips in voltage closer to the end of day. We hypothesize that this pattern may be due to the peak hours at the end of workdays when most families return home and turn on their electrical appliances and lighting. This may lead to overloading of the transformer, hence the observed voltage dips.
Following these patterns that were observed from Figure 3, we took a closer look at the behavior of the voltage profile on an hour-of-day basis, as illustrated in Figure 4 below.
The observations in Figure 3 affirm the similarity in voltage trends between sensors connected to the same transformer. Furthermore, in most sites there is a dip in voltage towards the end of the day, which we hypothesize can be explained by transformer overload (as described above). The consistent voltage fluctuation across days has tangible consequences: these voltage dips often lead to damage of electrical appliances, with many people switching off appliances when these fluctuations happen to avoid costs of appliance repair. Spending almost 20% of time with low voltage levels is also detrimental as this may damage equipment, lowering the economic productivity of households.
We measured the number and duration of power outages in each of the 150 sites and produced the associated standard key performance indicators (KPIs): SAIDI (system average interruption duration index) and SAIFI (system average interruption frequency index).
Computations for each bar are done from the sensor measurements of the four sensors in site 2297. To determine if an outage has occurred in a site, at least two sensors in that site must lose power within a few seconds of each other (usually 90 seconds). For a week such as August 15, SAIDI is low (~ 11 hours) while SAIFI is high (> 15 interruptions) which means many short interruptions were experienced that week. On the other hand, for a week like July 4th, SAIDI is comparatively large (~28 hours) while SAIFI is ~2.5 interruptions, indicating few long outages that week. These graphs also show that for the two-month data collection period in this site, there was almost always an outage every week, with outage durations ranging from several minutes to 50 hours in one week.
To answer this question, the research team paired nLine sensor data with socio-economic surveys, on-the-ground analysis of LMPC procurement contracts and inspection reports, in-depth informational interviews with management-level staff from AfDB and WB, and independent construction auditing data.
Researchers found no significant difference in power outages or in voltage quality experienced by households between AfDB and WB funded sites.
Figure 6 (below) illustrates that the procurement processes employed by the WB did not appear to have led to a meaningful reduction in power outages or improvements in voltage quality over the 12-month sensor data collection period.
While variations in power quality were observed between households in a site, the research team did not find significantly better power quality at WB sites with stricter conditionality requirements.
Variations in power quality can however be a result of more general engineering channels. For example, households farther away from the central transformer may experience worse voltage quality than households connected closer to the transformer. Figure 7a (below) illustrates how a household’s distance from the central transformer — along the low-voltage electricity wire —may impact voltage quality. Additionally, as infrastructure ages, voltage quality may worsen with the passing of time — thus emphasizing the importance of high-quality construction. Figure 7b (below) illustrates the observed effect of of passage of time (~5 years) since LV network construction began in the LMCP sites and voltage quality measurements in that site.
Under further investigation, the research team found that the worsening voltage quality at distances further from the transformer is primarily due to the increasing number of connections between the central transformer and the household rather than the distance itself. In other words, the addition of customer connections along the LV network can worsen voltage quality for households located further away on the network, by “increasing consumption of electricity by meters connected more closely to the transformer.”
Figure 8 (below) illustrates a second view of the relationship between a household’s distance from the central transformer and average daily voltage experienced by the household. From the best fit plot, it can be seen that a household’s distance to the central transformer does not affect the average daily voltage experienced by that household. Even for households that are up to 1,400 meters (~0.87 miles) from a central transformer, the impedance of these long lines does not cause a significant drop in average voltage.
We are thankful for the opportunity to use our power monitoring data to support work focused on better understanding the impacts of electricity infrastructure procurement and construction conditionalities on power quality and reliability.
In July, myself (Margaret), Prof. Susanna Berkouwer (a lead author of the study), and Dr. Noah Klugman (nLine’s CEO) shared these sensor data insights with Kenya’s Energy and Petroleum Regulatory Authority (EPRA), Kenya Power and Lighting Company (KPLC), the Kenya Power Institute of Energy Studies and Research, and researchers at the Strathmore University Energy Research Centre (SERC). We discussed potential applications of our sensor data in both daily operations and big-picture planning. Given the flexibility in sensor data applications, ease of deployment, and ability to produce granular as well as aggregated grid KPIs, we are excited to continue discussions to understand how our data and related products can be adapted and operationalized to better serve the needs of regulators, utilities and researchers alike.
If you are looking to collect power quality and reliability data in a context that would be difficult to reach with existing sensing technologies, please reach out to us at firstname.lastname@example.org.