Wearable and Internet of Things (IoT) devices are becoming popular in several applications, such as health monitoring, wide area sensing, and digital agriculture. These devices are energy-constrained due to limited battery capacities. As such, IoT devices harvest energy from the environment and manage it to prolong operation of the system. Stochastic nature of ambient energy, coupled with small battery sizes may lead to insufficient energy for obtaining data from all sensors. As a result, sensors either have to be duty cycled or subsampled to meet the energy budget. However, machine learning (ML) models for these applications are typically trained with the assumption that data from all sensors are available, leading to loss in accuracy. To overcome this, we propose a novel approach that combines data imputation with energy management (EM). Data imputation aims to substitute missing data with appropriate values so that complete sensor data are available for application processing, while EM makes energy budget decisions on the devices. We use the energy budget to obtain complete data from as many sensors as possible and turn off other sensors instead of duty cycling all sensors. Then, we use a low-overhead imputation technique for unavailable sensors and use them in ML models. Evaluations with six diverse datasets show that the proposed EM with imputation approach achieves 25%–55% higher accuracy when compared to duty cycling or subsampling without using additional energy.

Internet of Things (IoT) devices have the potential to enable a wide range of applications, including smart health and agriculture. However, they are limited by their small battery capacities. Utilizing energy harvesting is a promising approach to augment the battery life of IoT devices. However, relying solely on harvested energy is insufficient due to the stochastic nature of ambient sources. Predicting and accounting for uncertainty in the energy harvest (EH) is critical for optimal energy management (EM) in wearable IoT devices. This article proposes a two-step uncertainty-aware EH prediction and management framework for wearable IoT devices. First, the framework employs an energy-efficient conformal prediction (CP) method to predict future EH and construct prediction intervals. Contrasting to prior CP approaches, we propose constructing the prediction intervals using a combination of residuals from previous hours and days. Second, the framework proposes a near-optimal EM approach that utilizes a rollout algorithm. The rollout algorithm efficiently simulates various energy allocation trajectories as a function of predicted EH bounds. Using results from the rollout, the proposed approach constructs energy allocation bounds that maximize application utility (quality of service) with a high probability. Evaluations using real-world energy data from ARAS and Mannheim datasets show that the proposed CP for EH prediction provides 93% coverage probability with an average width of 9.5 J and 1.9 J, respectively. Moreover, EM using the rollout algorithm provides energy allocation decisions that are within 1.9–2.9 J of the optimal with minimal overhead.

Internet of things (IoT) devices are popular in several high-impact applications such as mobile healthcare and digital agriculture. However, IoT devices have limited operating lifetime due to their small form factor. Harvesting energy from ambient sources is an effective method to supplement the battery. Energy harvesting necessitates development of energy management policies to manage the harvested energy. Designing optimal policies for energy management is challenging for two key reasons: (1) ambient energy sources are highly stochastic; therefore, energy management policies must consider the associated uncertainty; (2) energy management policies must consider future energy availability while making decisions to ensure that sufficient energy is available when there is no ambient energy. Prior approaches typically consider energy in the immediate future (e.g., 1 hour) and do not account for the uncertainty in future energy harvest. This article proposes novel machine learning and dynamic optimization-based approaches to handle the two challenges. Specifically, we first develop a novel set of features and use it in a low-power neural network architecture to predict future energy availability and uncertainty. The energy predictions and uncertainty are used in a dynamic optimization algorithm to optimally allocate the harvested energy. Experiments on solar energy data over 5 years from Golden, Colorado, show that the proposed energy prediction model achieves 3.4 J mean absolute error while having a coverage of 80%. Moreover, our energy management algorithm provides energy allocations that are within 2.5 J of an optimal Oracle with 2.65 mJ to 36.54 mJ of energy overhead.

Wearable devices are becoming increasingly popular for health and activity monitoring applications. These devices typically include small rechargeable batteries to improve user comfort. However, the small battery capacity leads to limited operating life, requiring frequent recharging. Recent research has proposed energy harvesting using light and user motion to improve the lifetime of wearable devices. Most energy harvesting approaches assume that the placement of the energy harvesting device and sensors required for health monitoring are the same. However, this assumption does not hold for several real-world applications. For example, motion energy harvesting using piezoelectric sensors is limited to the knees and elbows, while a sensor for heart rate monitoring must be placed on the chest for optimal performance. To address this challenge, we propose a novel dynamic energy management approach referred to as DIET for wearable health applications enabled by multiple sensors and energy harvesting devices. The key idea behind DIET is to harvest energy from multiple sources and optimally allocate it to each sensor using a lightweight optimization algorithm such that the overall utility for applications is maximized. Experiments on real-world data from four users over 30 days show that the DIET approach achieves utility within 10% of an offline Oracle.

Internet of Things (IoT) devices are becoming popular in a number of transformative applications, including smart health, digital agriculture, and wide area sensing. However, small battery capacities and the need for frequent battery replacements or recharging have hindered their widespread adoption. Energy harvesting (EH) and management present a promising opportunity to enable long-term recharge free operation of IoT devices. State-of-the-art energy management approaches employ dynamic optimization methods to manage the harvested energy. However, the dynamic optimization methods are typically computationally intensive and lead to significant energy overhead for energy-constrained IoT devices. In contrast, this article proposes an imitation learning (IL)-based energy management algorithm that provides the energy budget or allocation for each decision interval without the need for dynamic optimization. We present an efficient approach to design Oracle policies that optimize the energy allocation of the IoT device to enable self-powered operation while maximizing the utility to the application. Then, we leverage the Oracle policy to train an online policy that performs near-optimal energy allocation at runtime. Our experiments with solar EH data for six years from three locations show that the proposed IL policies achieve allocation that is, on average, within 2.5 J of the Oracle, while having an energy consumption overhead of 154$\mu \text{J}$.

Low-power internet of things devices have the potential to transform multiple fields including healthcare, environmental monitoring, and digital agriculture. However, the operating life of these devices is severely constrained by their small batteries that require frequent recharging. Harvesting energy from ambient sources has emerged as an effective approach to prolong the lifetime of these devices. The harvested energy must be carefully managed to ensure that sufficient energy is available when ambient energy is scarce. Prediction of the energy available in the future can aid energy management algorithms in making better decisions about the allocation of the available energy. This paper proposes a novel hierarchical machine learning model that considers recent history and daily variations to make accurate predictions of future energy availability. We also propose using online learning to adapt the model to seasonal and spatial variations in the harvested energy. Using real-world solar energy data from the National Renewable Energy Laboratory we show that the proposed approach has mean absolute errors of 1.5 J and 1.1 J in predictions under same and different locations, respectively. We also demonstrate that using the proposed approach in an energy management algorithm leads to 54% higher utility and 95% lower battery violations than the baseline.

Autonomous electric mobility on demand (AEMoD) services are anticipated to be the future of private transportation, serving tens-of-thousands of requests per minute in large cities. To cope with this massive demand, a decentralized (i.e., zone-based) and multi-class management framework of AEMoD fleets was recently introduced. Yet, the inter-zone management of such approach has not been investigated. This paper thus fills this gap by studying the fleet re-balancing problem, with possible in-route charging, in decentralized multiclass AEMoD systems. A queuing model for multi-class re-balancing and possible in-route charging is developed on top of the system's decentralized fleet management. The stability conditions of this model are first derived, then the optimal inter-zone multi-class re-balancing and in-route charging decisions are derived so as to minimize the maximum response time in each deficient zone. Closed-form solutions are derived using Lagrangian analysis and simulations in a realistic setting in the city of Seattle are employed to illustrate the merits of our proposed re-balancing scheme as opposed to different baseline rebalancing approaches.