Last-mile consumers hugely inflate expected energy use, ballooning system costs (by 4 times in Kenya sample). Using aggregated data is the best forecasting tool, but mini-grid companies need to share to scale
Renewable energy powered mini-grids can provide clean, sustainable energy to the millions of people around the world who currently live without electricity access. A key factor to determining if a mini-grid is financially viable is the appropriate sizing of the mini-grid to the community’s electricity need. Each watt-hour (Wh) of over-predicted electricity demand can cost up to $6 of surplus equipment costs.
A widely employed method to estimate demand is through an energy use field survey; potential customers are asked which appliances they possess and plan to possess if they are connected to the mini-grid. Default appliance rating (in watts) and hours of use are used to estimate a daily consumption requirement.
In an analysis published in a recent paper, Dr. Henry Louie (Seattle University), Peter Dauenhauer (U of Strathclyde), and I used data from 8 solar mini-grids in Kenya owned by Vulcan Inc. to look at the accuracy of this demand prediction method. The research was based on responses from appliance surveys of customers conducted before the installation of the mini-grids, actual measured consumption over a 31-month period, and a follow-up appliance audit of a subset of the customers.
Measured consumption of electricity across the 8 mini-grids was an average of 113 Wh/customer/day, One mini-grid site (Entesopia) drives the average up, with the highest average of 230 Wh/customer/day. The top four appliances possessed by customers were lights, TVs, phone chargers, and radios.
We compared actual consumption to estimates from the appliance survey and appliance audit. In addition, we used actual consumption data to create a proxy approach under the assumption that the average consumption of one mini-grid should provide an accurate prediction of another grid. The average error (demand estimation compared to actual consumption) and potential extra cost from overestimation of demand for each method can be seen in the table below.
Potential causes for error include: generic survey process errors such as interviewer bias and satisfying behavior; appliance rating variance; and inaccurate estimations (assumed and from survey responses) for duration of appliance use. As can be seen, the most accurate method for calculating demand was to use aggregated data, making the case for the importance of sharing data. The level of inaccuracy of the survey approach is striking: very few customers were able to correctly predict their future consumption to even within +/- 25% of actual consumption (see figure).
When people use electricity is important for understanding how to size energy storage: night users require energy storage while day users can use solar electricity as it is generated. As seen below, we generated load profiles of a subset of customers and, in the paper, highlight a case study for each type of user.
43% of users are “Night Users” and consume 75% of energy between 18:00 – 6:00;
17% are “Day Users” and consume 50% of energy between 6:00 – 18:00;
40% are “Mixed Users” consume most energy during the night but also consume some energy during the day.
This research demonstrates that the commonly used appliance survey method for estimating customer demand provides poor predictions, resulting in a system costing four-times the amount system would have cost if it had been sized to meet actual demand. A data-driven approach is a more accurate alternative method of demand prediction.
Courtney Blodgett, Peter Dauenhauer, Henry Louie, Lauren Kickham, Accuracy of energy-use surveys in predicting rural mini-grid user consumption, In Energy for Sustainable Development, Volume 41, 2017, Pages 88-105, ISSN 0973-0826, https://doi.org/10.1016/j.esd.2017.08.002