Context-Aware Management of IoT Nodes: Balancing Informational Value with Energy Usage

TL;DR

This work addresses sustaining energy-harvesting IoT sensor nodes while supporting time-critical data collection. It introduces a context-aware energy management approach based on Model Predictive Control (MPC) that jointly optimizes sampling and transmission frequencies by maximizing a utility combining Value of Information (VoI) and State of Energy (SoE) over a finite horizon $H_p$, using a novel VoI formulation that captures threat proximity, process fidelity, and update delay. The proposed framework demonstrates adaptive behavior during real-world flash flood events, balancing information value with energy availability and outperforming fixed-duty-cycle baselines in VoI while avoiding energy depletion. The approach offers practical impact for remote, energy-constrained sensing where timely data is essential, enabling longer-lived networks without manual intervention.

Abstract

The operational lifetime of energy-harvesting wireless sensor nodes is limited by availability of the energy source and the capacity of the installed energy buffer. When a sensor node depletes its energy reserves, manual intervention is often required to resume node operation. While lowering the duty cycle would help extend the network lifetime, this is often undesirable, especially in time-critical applications, where rapid collection and dissemination of information is vital. In this paper, we propose a context-aware energy management policy that helps balance the two opposing objectives of timely data collection and dissemination with energy conservation. We capture these objectives through the Value of Information (VoI) of observations made by a sensor node and the State of Energy (SoE) of the energy buffer. We formulate the energy management policy as a Model Predictive Control (MPC) problem which computes device sampling and transmission frequencies to maximize a defined utility criterion over a finite, receding, time-horizon. In the process, we also develop a unique mathematical representation for VoI, that adequately captures aspects related to continuity in monitoring, urgency of dissemination, and representation of the phenomena being observed. In the end, we use data collected from a real-world flash flood event, to evaluate our decision framework across multiple scenarios of energy availability.

Figures (6)

• Figure 1: Timing diagram with the prediction horizon $H_p$. The dashed arrows represent the old decisions before updating the beliefs while the solid arrows are the decisions made after updating beliefs and are actually followed.
• Figure 2: Our system model depicting the flow of contextual information to the energy management policy. The selected frequencies in turn affect the projected State of Energy and Value of Information.
• Figure 3: Behavior of a risk-inclined and a risk-averse planner with changing stream levels. The parameters for the risk-inclined planner are $\lambda_c = 1$, $\alpha_r=0.009$, $\alpha_d=0.025$, $D_o=0.5$ and for the risk-averse planner are $\lambda_c = 0.5$, $\alpha_r=0.02$, $\alpha_d=0.25$, $D_o=0.25$
• Figure 4: The representative discharge profile. Note the short switching delay between sensors to stabilize before moving to the next sensing stage.
• Figure 5: Control actions, stream levels, and SoE for different initial SoE values. Both VoI and SoE are equally weighted i.e. $w_v = w_e = 0.5$