In low-income settings, the most critical piece of information for electric utilities is the anticipated consumption of a customer. Electricity consumption assessment is difficult to do in settings where a significant fraction of households do not yet have an electricity connection. In such settings the absolute levels of anticipated consumption can range from 5-100 kWh/month, leading to high variability amongst these customers. Precious resources are at stake if a significant fraction of low consumers are connected over those with higher consumption. This is the first study of it’s kind in low-income settings that attempts to predict a building’s consumption and not that of an aggregate administrative area. We train a Convolutional Neural Network (CNN) over pre-electrification daytime satellite imagery with a sample of utility bills from 20,000 geo-referenced electricity customers in Kenya (0.01% of Kenya’s residential customers). This is made possible with a two-stage approach that uses a novel building segmentation approach to leverage much larger volumes of no-cost satellite imagery to make the most of scarce and expensive customer data. Our method shows that competitive accuracies can be achieved at the building level, addressing the challenge of consumption variability. This work shows that the building’s characteristics and it’s surrounding context are both important in predicting consumption levels. We also evaluate the addition of lower resolution geospatial datasets into the training process, including nighttime lights and census-derived data. The results are already helping inform site selection and distribution-level planning, through granular predictions at the level of individual structures in Kenya and there is no reason this cannot be extended to other countries.

In much of the world, electricity grids are not instrumented at the customer level, limiting insights into the power quality experienced by utility customers. Moreover, to understand grid performance, regulators and investors must depend on utilities to self-report reliability data. To address these challenges, we introduce PowerWatch, an agile methodology to directly measure customer experience and aggregated grid performance without relying on the utility for deployment or management. PowerWatch employs a system of distributed sensors coupled with cloud-based analytics. We evaluate the PowerWatch methodology by deploying 462 sensors in homes and businesses in Accra, Ghana for over a year, yielding the largest open-source data set on electricity reliability at the customer-level in the region. We describe the architecture, design, and performance of PowerWatch, as well as the data that are collected, explaining how we determine the accuracy and coverage of our methodology without ground truth. Finally, we report on grid performance issues, finding nearly twice as many outages as the utility observed, suggesting a need for better grid performance monitoring.

The means of electrifying households and the resulting electricity networks are rapidly evolving. Traditionally, an extension of existing centralized grids was the only prominent technique, but now electrification is seeing massive expansion via decentralized solar home systems (SHSs). These systems consist of a low-wattage photovoltaic (PV) panel (typically 5-100W), a battery, a collection of energy-efficient DC appliances, and a charge controller. Spurred by significant advances and reduced costs in solar, batteries, energyefficient appliances, and mobile money-driven business models, SHSs have proliferated rapidly, with tens of millions of systems now deployed, primarily in regions with otherwise low rates of electricity access.

To extract information at scale, researchers are increasingly applying semantic segmentation techniques to remotely-sensed imagery. While fully-supervised learning enables accurate pixelwise segmentation, compiling the exhaustive datasets required is often prohibitively expensive, and open-source datasets that do exists are frequently inexact and non-exhaustive. In this paper, we present a novel and generalizable two-stage framework that enables improved pixelwise image segmentation given misaligned and missing annotations. First, we introduce the Alignment Correction Network to rectify incorrectly registered open source labels. Next, we demonstrate a segmentation model - the Pointer Segmentation Network - that uses corrected labels to predict infrastructure footprints despite missing annotations. We demonstrate the transferability of our method to lower quality data sources by applying the Alignment Correction Network to correct OpenStreetMaps building footprints, and we show the accuracy of the Pointer Segmentation Network in predicting cropland boundaries in California. Overall, our methodology is robust for multiple applications with varied amounts of training data present, thus offering a method to extract reliable information from noisy, partial data.

High-quality roads are the scaffolding for prosperous and healthy societies, and accordingly garner huge investments from governments every year. However, current techniques to monitor those investments tend to be time-consuming, laborious, and expensive, placing them out of reach for many developing regions. In this work, we develop a model for monitoring the quality of road infrastructure using satellite imagery, enabling much larger scale and much lower costs than are achievable with current methods. For this task, we harness two trends: the increasing availability of high-resolution, often-updated satellite imagery, and substantial improvement in accuracy and performance of neural network-based methods for executing computer vision tasks. In this study, we train a model for intercity road quality prediction using a unique dataset of road quality measurement labels (57 roads, total length is 7000km) throughout the Republic of Kenya combined with corresponding 50cm resolution satellite imagery. Using a variety of neural network architectures, we create and evaluate regression models for predicting road quality. Our results show a best-case R2 value of 0.79 for the regression problem using a standard train-test split and an R2 value of 0.35 for the substantially harder held-out regression problem which has the added potential to generalize more readily to other contexts. We further demonstrate the potential of our measurement technique with a large-scale case study (for 322 towns throughout Kenya) that shows a positive relationship between incoming road quality and nighttime illumination, a common proxy measurement for local economic activity. These results indicate the possibility to measure road quality at an unprecedented scale, providing insight into the contribution of high-quality roads to many societal development indicators.

The vision of sensor systems that collect critical and previously ungathered information about the world is often only realized when sensors, students, and subjects move outside the academic laboratory. However, deployments at even the smallest scales introduce complexities and risks that can be difficult for a research team to anticipate. Over the past year, our interdisciplinary team of engineers and economists has been designing, deploying, and operating a large sensor network in Accra, Ghana that measures power outages and quality at households and firms. This network consists of 457 custom sensors, over 3,000 mobile app instances, thousands of participant surveys, and custom user incentive and deployment management systems. In part, this deployment supports an evaluation of the impacts of investments in the grid on reliability and the subsequent effects of improvements in reliability on socioeconomic well-being. We report our experiences as we move from performing small pilot deployments to our current scale, attempting to identify the pain points at each stage of the deployment. Finally, we extract high-level observations and lessons learned from our deployment activities, which we wish we had originally known when forecasting budgets, human resources, and project timelines. These insights will be critical as we look toward scaling our deployment to the entire city of Accra and beyond, and we hope that they will encourage and support other researchers looking to measure highly granular information about our world's critical systems.

Incentives are a key facet of human studies research, yet the state-of-the-art often designs and implements incentive systems in an ad-hoc, on-demand manner. We introduce the first vocabulary for formally describing incentive systems and develop a software infrastructure that enables UI-based graphical generation of complex, auditable, reliable, and reproducible incentive systems. We call this infrastructure the Open INcentive Kit (OINK). A review of recent literature from several communities finds that of the one hundred and twenty-one publications that incorporate incentives, only thirty-one describe their incentive system in detail, and all of these could be implemented using OINK. We evaluate OINK in practice by using it for an active energy monitoring deployment in Ghana and find that OINK successfully facilitates thousands of individual incentive payments. Finally, we describe our efforts to generalize OINK for different research communities, specifically focusing on architectural decisions around extensibility to support unanticipated use cases. OINK is free and open-source software.

Power systems have evolved following a century-old paradigm of planning and operating a grid based on large central generation plants connected to load centers through a transmission grid and distribution lines with radial flows. This paradigm is being challenged by the development and diffusion of modular generation and storage technologies. We use a novel approach to assess the sequencing and pacing of centralized, distributed, and off-grid electrification strategies by developing and employing the grid and access planning (GAP) model. GAP is a capacity expansion model to jointly assess operation and investment in utility-scale generation, transmission, distribution, and demand-side resources. This paper conceptually studies the investment and operation decisions for a power system with and without distributed resources. Contrary to the current practice, we find hybrid systems that pair grid connections with distributed energy resources (DERs) are the preferred mode of electricity supply for greenfield expansion under conservative reductions in photovoltaic panel (PV) and energy storage prices. We also find that when distributed PV and storage are employed in power system expansion, there are savings of 15%-20% mostly in capital deferment and reduced diesel use. Results show that enhanced financing mechanisms for DER PV and storage could enable 50%-60% of additional deployment and save 15 $/MWh in system costs. These results have important implications to reform current utility business models in developed power systems and to guide the development of electrification strategies in underdeveloped grids.

We describe the design, development, and characterization of a system that uses only a commodity smartphone and a consumer fan to measure the voltage of an electricity grid. This system represents the first known example of a system for measuring power quality of an electricity grid without specialized equipment, and can enable those with electricity access who live in "weak grid" environments to determine the current state of their electricity system, especially whether it is safe to plug in appliances. Our work leverages the microphone of the phone to measure the disturbance created on a pure sinusoidal tone into a running fan. The distance between adjacent harmonics in this disturbance is a direct proxy for the electricity grid voltage. We build a system to exploit this phenomenon and show that it is robust to a variety of fan types, phone types, and background audio environments, consistently producing voltage estimations within +/- 2%. We also show that our method works for smartphones as well as feature phones, making it applicable in a wide range of settings.

Network connectivity is often one of the most challenging aspects of deploying sensors. In many countries, cellular networks provide the most reliable, highest bandwidth, and greatest coverage option for Internet access. While this makes smartphones a seemingly ideal platform to serve as a gateway between sensors and the cloud, we find that a device designed for multi-tenant operation and frequent human interaction becomes unreliable when tasked to continuously run a single application with no human interaction, a seemingly counter-intuitive result. Further, we find that economy phones cannot physically withstand continuous operation, resulting in a surprisingly high rate of permanent device failures in the field. If these observations hold more broadly, they would make mobile phones poorly suited to a range of sensing applications for which they have been rumored to hold great promise.

The rapid expansion of intermittent grid-tied solar capacity is making the job of balancing electricity’s real-time supply and demand increasingly challenging. Recent work proposes mechanisms for actively controlling solar power in the grid at individual sites by enabling software to cap it as a fraction of its time-varying maximum output. However, while enforcing an equal fraction of each solar site’s time-varying maximum output results in “fair” short-term contributions of solar power across all sites, it does not result in “fair” long-term contributions of solar energy. Enforcing fair long-term energy access is important when controlling distributed solar capacity, since limits on solar output impact the compensation users receive for net metering and the battery capacity required to store excess solar energy. This discrepancy arises from fundamental differences in enforcing “fair” access to the grid to contribute solar energy, compared to analogous fair sharing in networks and processors. To address the problem, we first present both a centralized and distributed algorithm to enable control of distributed solar capacity that enforces fair grid energy access. We then present multiple policies that show how utilities can leverage this new distributed rate-limiting mechanism to reduce variations in grid demand from intermittent solar generation.

We present insights obtained from conducting a year-long, 455 meter deployment of wireless plug-load electric meters in a large commercial building. We develop a stratified sampling methodology for surveying the energy use of Miscellaneous Electric Loads (MELs) in commercial buildings, and apply it to our study building. Over the deployment period, we collected over nine hundred million individual readings. Among our findings, we document the need for a dynamic, scalable IPv6 routing protocol which supports point-to-point routing and multiple points of egress. Although the meters are static physically, we find that the set of links they use is dynamic; not using such a dynamic set results in paths that are twice as long. Finally, we conduct a detailed survey of the accuracy possible with inexpensive AC metering hardware. Based on a 21-point automated calibration of a population of 500 devices, we find that it is possible to produce nearly utility-grade metering data.

Heating, ventilation, and air conditioning (HVAC) systems are an important target for efficiency improvements through new equipment and retrofitting because of their large energy footprint. One type of equipment that is common in homes and some offices is an electrical, single-stage heat pump air conditioner (AC). To study this setup, we have built the Berkeley Retrofitted and Inexpensive HVAC Testbed for Energy Efficiency (BRITE) platform. This platform allows us to actuate an AC unit that controls the room temperature of a computer laboratory on the Berkeley campus that is actively used by students, while sensors record room temperature and AC energy consumption. We build a mathematical model of the temperature dynamics of the room, and combining this model with statistical methods allows us to compute the heating load due to occupants and equipment using only a single temperature sensor. Next, we implement a control strategy that uses learning-based model-predictive control (MPC) to learn and compensate for the amount of heating due to occupancy as it varies throughout the day and year. Experiments on BRITE show that our techniques result in a 30%-70% reduction in energy consumption as compared to two-position control, while still maintaining a comfortable room temperature. The energy savings are due to our control scheme compensating for varying occupancy, while considering the transient and steady state electrical consumption of the AC. Our techniques can likely be generalized to other HVAC systems while still maintaining these energy saving features.

We describe the design, deployment, and experience with a wireless sensor network for high-fidelity monitoring of electrical usage in buildings. A network of 38 mote-class AC meters, 6 light sensors, and 1 vibration sensor is used to determine and audit the energy envelope of an active laboratory. Classic WSN issues of coverage, aggregation, sampling, and inference are shown to appear in a novel form in this context. The fundamental structuring principle is the underlying load tree, and a variety of techniques are described to disambiguate loads within this structure. Utilizing contextual metadata, this information is recomposed in terms of its spatial, functional, and individual projections. This suggests a path to broad use of WSN technology in energy and environmental domains.

This paper describes a systematic approach to building micro-solar power subsystems for wireless sensor network nodes. Our approach composes models of the basic pieces - solar panels, regulators, energy storage elements, and application loads - to appropriately select and size the components. We demonstrate our approach in the context of a microclimate monitoring project through the design of the node, micro-solar subsystem, and network, which is deployed in a challenging, deep forest setting. We evaluate our deployment by analyzing the effects of the range of solar profiles experienced across the network.

We present a building block approach to hardware platform design based on a decade of collective experience in this area, arriving at an architecture in which general-purpose modules that require expertise to de sign and incorporate commonly-used functionality are integrated with application-specific carriers that satisfy the unique sensing, power supply, and mechanical constraints of an application. Of course, modules are widespread, but our focus is far less on the performance of any individual module and far more on an overall architecture that supports the prototype, pilot, and production stages of design, and preserves the artifacts and learnings accumulated along the way.

A home where every major appliance can be monitored for energy consumption and individually controlled wirelessly has long been a dream of gadgeteers and the green-conscious alike. Research has shown that real-time, per-appliance electricity usage feedback can induce behavior changes that lead to 10% to 20% reduction in usage [2].

We present the philosophy, design, and initial evaluation of the Trio Testbed, a new outdoor sensor network deployment that consists of 557 solar-powered motes, seven gateway nodes, and a root server. The testbed covers an area of approximately 50,000 square meters and was in continuous operation during the last four months of 2005. This new testbed in one of the largest solar-powered outdoor sensor networks ever constructed and it offers a unique platform on which both systems and application software can be tested safely at scale. The testbed is based on Trio, a new mote platform that provides sustainable operation, enables efficient in situ interaction, and supports fail-safe programming. The motivation behind this testbed was to evaluate robust multi-target tracking algorithms at scale. However, using the testbed has stressed the system software, networking protocols, and management tools in ways that have exposed subtle but serious weaknesses that were never discovered using indoor testbeds or smaller deployments. We have been iteratively improving our support software, with the eventual aim of creating a stable hardware-software platform for sustainable, scalable, and flexible testbed deployments.

A main challenge with developing applications for wireless embedded systems is the lack of visibility and control during execution of an application. In this paper, we present a tool suite called Marionette that provides the ability to call functions and to read or write variables on pre-compiled, embedded programs at run-time, without requiring the programmer to add any special code to the application. This rich interface facilitates interactive development and debugging at minimal cost to the node.