Emedia: ISSN 1529-7306 Emedia/Online Inc.

Emedia: ISSN 1529-7306 Vol.39_1, 1-18

Commons Attribution 4.0 International License, which permits unrestricted use, distribution, and

reproduction in any medium, provided the original work is properly cited.

An Adaptive Task Scheduling Based Energy Optimization

Model for IoT Devices

Lingli Yao*

Nanjing Vocational College of Information Technology, Nanjing, 210046

*Corresponding Author:Lingli Yao

Email: yaoll@njcit.cn

Received 10 April 2026; Accepted 15 July 2026

Abstract：This research presents a novel energy optimization model for Internet of Things

(IoT) devices by integrating adaptive task scheduling algorithms with dynamic energy

management mechanisms to enhance energy efficiency in IoT systems. We develop an

innovative Dual-Factor Adaptive Efficiency Function (DFAEF) that comprehensively

considers both device workload and battery status to achieve precise efficiency evaluation. The

model employs a multi-parameter weighted scoring mechanism for task allocation and

enhances overall system performance through network topology optimization. Simulation

results demonstrate that, compared to traditional methods, the proposed model reduces energy

consumption by approximately 31% while maintaining high task distribution balance and

network connectivity. This research provides IoT system designers with a practical energy

optimization solution that offers significant value for extending battery-powered device

lifespan and enabling sustainable operations.

Keywords: Internet of Things; energy optimization; task scheduling; Dual-Factor Adaptive

Efficiency Function; network topology

1. Introduction

The flourishing development of Internet of Things (IoT) technology has profoundly impacted

multiple domains including industrial automation, smart cities, and environmental monitoring.

With global IoT device numbers projected to reach 75 billion by 2025 [1,2], energy

management has emerged as a critical challenge for ensuring sustainable development of IoT

systems. Battery-powered IoT devices face stringent energy constraints that affect their

operational longevity and system reliability[3,4]. Recent research by Almudayni.[5] et al

indicates that energy efficiency not only influences device lifespan but also directly determines

deployment costs and maintenance cycles, particularly in environments with limited

accessibility. Consequently, optimizing energy consumption in IoT systems has become a

focal direction for research.

Existing studies explore IoT energy optimization from three primary dimensions: hardware

layer, network protocols, and system management. At the hardware level, heterogeneous

Lingli Yao

multi-core architectures proposed by Liu[6] et al. and adaptive voltage regulation techniques

developed by Wang[7] et al. have significantly reduced processor energy consumption.

Regarding network protocols, Zhang[8] et al.'s optimized routing model and Pereira[9] et al.

network interface average power metric have effectively extended network lifetime. In system

management, Gupta[10] et al. designed a dynamic homomorphic security scheme that

optimizes energy utilization while ensuring data security. However, these studies often focus

on single dimensions, lacking comprehensive optimization strategies that consider device

heterogeneity, dynamic workloads, and network topology.

To address these limitations, this paper proposes an adaptive task scheduling based energy

optimization model for IoT devices with four key innovations. First, it designs a Dual-Factor

Adaptive Efficiency Function (DFAEF) that comprehensively considers device workload and

battery status, enabling precise evaluation and prediction of device energy efficiency. Second,

it develops an intelligent task allocation mechanism based on multi-parameter weighting that

achieves dynamic balance among computational capability, network bandwidth, energy

efficiency, and battery status. Third, it designs an adaptive power allocation algorithm that

responds to variations in device workload and network conditions. Fourth, it proposes a

network topology optimization method based on device characteristics that significantly

enhances overall communication efficiency of the system.

The main contributions of this research include: the first multi-dimensional IoT energy

optimization framework that simultaneously considers device characteristics, task

requirements, and network structure; development of adaptive efficiency functions and task

scheduling algorithms capable of adapting to dynamic environments; verification through

large-scale simulation demonstrating that the proposed model reduces energy consumption by

31.2% while maintaining high-performance system operation; provision of a set of energy

optimization strategies that can be extended to various IoT application scenarios. Experimental

results indicate that the proposed model performs excellently in multiple application scenarios

including smart cities, industrial monitoring, and environmental sensing, offering an effective

solution to energy management challenges in IoT systems.

2. Related Work

Energy optimization in IoT systems has been a research hotspot. This section reviews relevant

research from three aspects: hardware design, network protocols, and system management.

2.1 Low-Power Hardware Design

At the hardware level, low-power design is key to reducing IoT device energy consumption.

Liu[6] et al. proposed an IoT device architecture based on heterogeneous multi - core

processors for energy optimization via dynamic core frequency adjustment. Wang[11] et al.

developed an adaptive voltage scaling technique to adjust processor voltage according to

workload, cutting energy use. Rahman[12] et al. designed a low - power sensor network

An Adaptive Task Scheduling Based Energy Optimization Model for IoT Devices

using sleep mechanisms to save energy. However, these methods need specific hardware,

limiting their universality.

2.2 Energy - Aware Network Protocols

Network communication is a major energy consumer in IoT devices. In recent years, various

energy - aware network protocols have been proposed. Chen[13] et al. introduced the E -

AOMDV routing protocol to adjust strategies based on node energy. Kim [14] et al. developed

a low - power MAC protocol to save energy by reducing unnecessary operations. Zhang[15] et

al. proposed a clustering - based data aggregation method to cut network data volume and

energy use. But these methods mainly focus on protocol optimization and ignore

computational task allocation.

2.3 System - Level Energy Management

System - level energy management aims to optimize IoT system energy consumption

holistically. Li [16] et al. proposed an energy management framework using deep

reinforcement learning for optimal strategies. Singh[17] et al. developed a task offloading

mechanism based on device energy and network conditions. Anderson[18] et al. proposed an

energy-aware scheduling algorithm to optimize task timing by predicting energy consumption.

Although progress has been made in each field, these studies mainly focus on single-aspect

optimization and lack comprehensive consideration of multi- dimensional factors in IoT

systems. The model in this paper fills this gap by integrating device characteristics, workloads,

network topology, etc. for more comprehensive energy optimization.

3. System Model and Problem Definition

3.1 System Architecture

We consider a system comprising N heterogeneous IoT devices, each with different

computational capabilities, network bandwidth, battery capacity, and energy consumption

characteristics. These devices form a network through wireless communication to collectively

complete a series of tasks. The primary objective of the system is to minimize overall energy

consumption while ensuring tasks are processed in a timely manner.

The set of all devices in the system can be mathematically represented as:

D={d1,d2,…,dN}⊆ℝ7 （1）

The topological structure of the network can be represented as a weighted directed graph:

Lingli Yao

G=(V,E,W) （2）

where V={v1,v2,…,vN} represents the set of vertices corresponding to the devices, E⊆V×V

represents the set of edges corresponding to communication links between devices, and

W:E→ℝ+ is a weight function that assigns a positive real number to each edge, representing

the energy consumption of communication between the corresponding devices. The weight

function can be defined as:

W(vi,vj)=ω1⊆dist(vi,vj)2+ω2⊆1

+ω3⊆1

Bj （3）

where dist(vi,vj) represents the Euclidean distance between devices di and dj, ω1, ω2, and ω3

are weighting coefficients, and Bi and Bj represent the bandwidth of devices di and dj,

respectively.

Figure 1. IoT System Network Topology Structure.

Figure 1 shows the distribution and connections of different device types in the system. Blue

nodes (42%) are edge sensors for data collection, red nodes (28%) are relays for data

forwarding, green nodes (18%) are gateways for network connection, and orange nodes (12%)

are servers for complex computations. Node size represents device computational capability

and importance, and lines are communication links. By using the K - means clustering

An Adaptive Task Scheduling Based Energy Optimization Model for IoT Devices

algorithm, device types are automatically identified according to connectivity and

computational capability, thus forming an efficient hierarchical network.

3.2 Device Model

Each IoT device di can be represented as a seven-tuple:

di=(Pi

base,Ci,Bi,Ei

cap,Ei

cur,Pri,Li) （4）

In this model, Pi

base denotes the device's baseline power consumption (mW), Ci represents its

computational capability (relative units), Bi denotes its network bandwidth (Mbps), Ei

cap

represents its battery capacity (mAh), Ei

cur indicates the current battery level (percentage), Pri

represents device priority (0 - 10), and Li indicates the current load (0 - 1).

The computational capability of a device can be further decomposed into multiple factors:

Ci=α1⊆fi

cpu+α2⊆Mi+α3⊆Ni

cores+α4⊆Si

cache （5）

where fi

cpu is the CPU frequency (GHz), Mi is the memory capacity (GB), Ni

cores is the number

of CPU cores, Si

cache is the cache size (MB), and α1, α2, α3, and α4 are weighting coefficients.

The battery energy dynamics of a device can be modeled using a first-order differential

equation:

dEi

cur(t)

dt =− Pi(t)⊆Vi

cap⊆3600 ⊆100% （6）

Energy consumption is calculated as followsE=P×t. For unit conversions, use mW for power

Pi(t)and s tfor time. The energy formula is 1 J = 1 mW × 1000 s = 0.278 mWh. The factor

3600 converts hours to seconds, and 100% converts values to percentage format.IoT devices

typically have a battery voltage Vi of 3.3 V or 3.7 V.The equation describes the battery - level

change rate over time,Pi(t) involving device power consumption diat a timtVie, operating

voltage, and the 3600 conversion factor for hours to seconds. Through this seven - tuple

modeling approach, we can comprehensively describe IoT devices' energy consumption

characteristics and operational status, providing a basis for subsequent task scheduling and

energy optimization.

3.3 Task Model

The set of tasks to be processed by the system is represented as T={t1,t2,...,tM}, with each task

tj defined as:

Lingli Yao

tj=(Cj

req,Dj

req,Prj

task,τj

arr,τj

ddl,ξj,Γj) （7）

Here, Cj

req represents the computational requirements of the task (relative units), reflecting the

computational complexity and resource demands of the task; Dj

req represents the data

transmission requirements of the task (relative units), indicating the data scale and

transmission volume involved in the task; Prj

task represents task priority (1-10), used to

distinguish the urgency and importance of tasks; τj

arr is the arrival time of the task; τj

ddl is the

deadline by which the task must be completed; ξj is the task complexity factor; and Γj is the set

of predecessor tasks that must be completed before task tj can start execution.

The computational requirement Cj

req of a task can be more precisely defined as:

req=ξj⊆log2(1+δj)⊆√Ij

data⊆Oj

data （8）

where ξj is the task complexity factor, δj is the difficulty level of the algorithm used, Ij

data is the

input data size (KB), and Oj

data is the output data size (KB).

Similarly, the data transmission requirement Dj

req can be formulated as:

req=Ij

data+Oj

data+ρj⊆∑Ok

data

tk⊆Γj （9）

where ρj is a coefficient indicating the proportion of predecessor tasks' output data required for

the current task.

The task model encompasses multiple key dimensions—computation, communication, priority,

time constraints, and dependencies—capable of accurately characterizing the diverse features

of tasks in IoT environments. By analyzing these attributes of tasks, the system can more

precisely evaluate task resource requirements and make optimal scheduling decisions

accordingly.

Information Gain-Based Feature Weight Optimization

To address the "black box" issue of feature weights, we employ an information gain-based

feature weight optimization method. For the feature matrix Fi=[degi,Ci,Ei

cap,Pi

base,centri]T,

weights are determined as:

wk=IG(S,Ak)

∑I

j=1 G(S,Aj) （10）

An Adaptive Task Scheduling Based Energy Optimization Model for IoT Devices

where IG(S,Ak) is the information gain of the k-th feature for device set S.

Table 1.Optimized Feature Weights (validated through experiments)

Feature Name

Symbol

Weight

Impact Description

Computational capability

0.35

Greatest impact on server classification

Battery capacity

cap

0.25

Determines device endurance

Centrality

centri

0.20

Affects network connectivity

Base power consumption

base

0.15

Key energy efficiency indicator

Degree

degi

0.05

Minimal impact

Weighted Clustering Distance:

d(Fi,Ck)=

√∑

j=1 ⊆(Fi,j−Ck,j)2 （11）

4.Proposed Energy Optimization Model

Based on the system architecture, device model, and task model defined in Section 3, this

section formulates and designs an adaptive energy optimization framework for IoT devices.

The proposed framework takes the system-wide energy minimization under task timeliness

constraints as the primary objective, and further incorporates device heterogeneity, dynamic

workloads, and network topology characteristics into the optimization process.

4.1 Dual-Factor Adaptive Efficiency Function (DFAEF)

To precisely evaluate device energy efficiency, we propose the "Dual-Factor Adaptive

Efficiency Function (DFAEF)," which comprehensively considers device workload and battery

level:

η(L,E)=e−k1(L−Lopt)2⊆(1−e−k2⊆E)⊆α （12）

where L represents device workload (within the 0-1 range), E represents battery level (within

the 0-1 range), Lopt is the optimal workload level (set to 0.65 based on empirical studies), k1 is

Lingli Yao

the workload influence coefficient, set to 4.0, k2 is the battery level influence coefficient, set to

0.8, and α is a model parameter used to control the overall weight of energy efficiency.

The partial derivatives of the efficiency function with respect to workload and battery level

provide insights into the sensitivity of the function to each factor:

∂η(L,E)

∂L=−2k1(L−Lopt)⊆e−k1(L−Lopt)2⊆(1−e−k2⊆E)⊆α （13）

∂η(L,E)

∂E=k2⊆e−k2⊆E⊆e−k1(L−Lopt)2⊆α （14）

The function has three important characteristics: efficiency increases with increasing workload

but reaches a peak at a threshold of approximately 0.65, after which it decreases with further

increases in workload, reflecting that devices typically achieve optimal efficiency under

moderate workload; lower battery levels correspond to lower efficiency, considering the need

for energy conservation in low battery states; by adjusting parameter α, the overall efficiency

level can be flexibly controlled to adapt to different application scenarios.

The second-order partial derivatives further reveal the rate of change of sensitivity:

∂2η(L,E)

∂L2=[−2k1+4k1

2(L−Lopt)2]⊆e−k1(L−Lopt)2⊆(1−e−k2⊆E)⊆α

∂2η(L,E)

∂E2=−k2

2⊆e−k2⊆E⊆e−k1(L−Lopt)2⊆α

∂2η(L,E)

∂L∂E=−2k1k2(L−Lopt)⊆e−k1(L−Lopt)2⊆e−k2⊆E⊆α

These mathematical properties ensure that DFAEF can comprehensively evaluate the actual

efficiency of devices under different workload and battery level conditions, providing accurate

decision-making basis for task scheduling.

An Adaptive Task Scheduling Based Energy Optimization Model for IoT Devices

Figure 2. Three-dimensional Visualization of the Dual-Factor Adaptive Efficiency

Function (DFAEF).

The horizontal axis represents device workload (0-1), the vertical axis represents battery level

(0-1), and the perpendicular axis represents the efficiency coefficient. Colors ranging from

blue to yellow indicate efficiency values from low to high. Color bar: Efficiency coefficient (0-

1); X-axis: Device workload (0-1); Y-axis: Battery level (0-1); Z-axis: DFAEF efficiency

value. The surface visualization demonstrates the non-linear relationship between workload,

battery status, and device efficiency, with optimal efficiency (yellow regions) occurring at

moderate workload levels (0.6-0.7) and high battery levels (0.8-1.0).

Figure 2 shows, three key observations can be made: (1) efficiency reaches a peak at a

workload of approximately 0.65, with both higher and lower workloads leading to decreased

efficiency; (2) higher battery levels correspond to higher efficiency, following a non-linear

relationship; (3) a high-efficiency plateau forms in the region of moderate workload (0.5-0.7)

and high battery level (>0.7), representing the optimal operational state range for the system.

4.2 Dynamic Power Allocation Algorithm

Based on the Dual-Factor Adaptive Efficiency Function (DFAEF) proposed in Section 4.1, this

section designs a dynamic power allocation algorithm that adjusts device power consumption

in real time according to workload and battery status：

Pi(t)= Pi

base⊆Li(t)

η(Li(t),Ei(t)) +Pi

comm(t)+Pi

idle(t) （15）

Lingli Yao

where Pi(t) represents the total power consumption of device di at time t, Li(t) and Ei(t)

represent the device's workload and battery level at time t, respectively. Picomm(t)

represents the communication power consumption, and Piidle(t) represents the idle

power consumption.

The communication power consumption can be modeled as:

To enhance the model's adaptability to real wireless environments, we introduce a

protocol adaptation layer and channel quality factors. The enhanced communication

energy model is:

comm(t)= ∑[αprotocol⊆Pi

tx⊆Dij

tx(t)

Bi⊆γretrans+βprotocol⊆Pi

rx⊆Dji

rx(t)

Bi]

j⊆Ni⊆γSNR

（16）

where the key parameters are defined as follows:

Protocol Adjustment Coefficients:

（1）LoRa protocol:

αprotocol=1.2+0.1×SF

where SF is the spreading factor .

（2）NB-IoT protocol:

βprotocol=0.9+0.2×sleep_ratio

considering sleep-wake mechanisms

（3）Wi-Fi protocol:

αprotocol=βprotocol=1.0 (baseline reference)

Channel Quality Factor:

An Adaptive Task Scheduling Based Energy Optimization Model for IoT Devices

γSNR=1+δ⊆e−SNR/10

Where δ=0.3 is the retransmission penalty coefficient and SNR is the signal-to-noise ratio.

Retransmission Impact Factor:

γretrans=1+ Nretrans

Ntotal ×0.5

Where Nretrans is the number of retransmissions and Ntotal is the total transmission attempts.

Table 2.Protocol Characteristics Analysis

Protocol

Power Range

Spreading Factor Impact

Sleep Power

Channel Occupancy

LoRa

14-20 dBm

Ptx⊆2SF/2

< 1 μA

Tair=PL/DR

NB-IoT

23 dBm (max)

Fixed 180kHz

PSM mode

Scheduling delay

Wi-Fi

15-20 dBm

Channel contention

Idle listening

CSMA/CA

Channel Interference Modeling

In dense IoT scenarios, channel conflicts significantly impact energy consumption. We

introduce channel occupancy rate ρ and collision probability Pc:

Pc=1−e−2ρτ （17）

The expected number of retransmissions is:

E[Nretrans]= Pc

1−Pc （18）

Where Ni is the set of neighboring devices of device di, Pi

tx and Pi

rx are the transmission and

reception power consumption rates, respectively, and Dij

tx(t) and Dji

rx(t) are the amounts of data

transmitted to and received from device dj at time t.

The idle power consumption follows an exponential decay model to simulate the power-

saving modes of IoT devices:

Lingli Yao

idle(t)=Pi

base⊆(1−Li(t))⊆e−λi⊆(1−Li(t)) （19）

where λi is a device-specific parameter controlling the rate of power reduction in low-

workload states.

The rate of change of battery level can now be expressed as a function of total power

consumption:

dEi(t)

dt =− Pi(t)⊆Vi

cap⊆3600 ⊆100% （20）

Where Vi is the operating voltage and Eicap is the battery capacity in mAh.

The time complexity of this algorithm is O(|T|⊆|D|⊆|Navg|), where |T| is the number of time

periods, |D| is the number of devices, and |Navg| is the average number of neighbors per device.

Figure 3. Power Consumption and Efficiency Analysis of IoT Devices.

(a) shows the average power consumption variation curve of IoT devices in a day. The

blue line is average power consumption, and the light blue area is the standard -

deviation range. Power consumption has clear periodic variation, being higher from 8:00

- 12:00 and 14:00 - 18:00 (work - time peaks) and dropping significantly at night,

showing the model's adaptive performance. (b) shows the performance variation of the

Dual - Factor Adaptive Efficiency Function in a day. The green line is average efficiency,

and the light green area is the standard - deviation range. The efficiency coefficient

averages 0.5-0.7, meaning devices operate in high - efficiency ranges mostly. Efficiency

decreases slightly with continuous battery consumption, but overall fluctuation is small,

demonstrating the model's long - term operational stability.

4.3 Multi-Parameter Weighted Task Scheduling Algorithm

An Adaptive Task Scheduling Based Energy Optimization Model for IoT Devices

Building on the DFAEF-based efficiency evaluation in Section 4.1 and the dynamic power

allocation mechanism in Section 4.2, this section presents a multi-parameter weighted

task scheduling algorithm that jointly considers computational capability,

communication bandwidth, energy efficiency, and battery status when assigning tasks to

devices.The algorithm first initializes the workload of all devices to random values

between 0.2 and 0.6, simulating initial state differences of devices in actual environments.

For each task to be allocated, the algorithm traverses all available devices and calculates

four key indicators, including computational capability, communication bandwidth,

energy efficiency, and current battery level. Subsequently, the algorithm computes each

device’s adaptation score for the current task using a comprehensive scoring formula and

adds a workload penalty term derived from the DFAEF function to discourage

overloading heavily utilized devices. After evaluating all devices, the algorithm selects the

device with the highest score as the optimal execution node for the current task, allocates

the task to that device, and updates the device’s workload status. Upon completing

allocation of all tasks, the algorithm returns the final task allocation scheme.

The time complexity of this algorithm is O(M⋅N), where M is the number of tasks and N is

the number of devices. The space complexity is O(M+N).

Figure 4. Task Allocation Results and Energy Consumption Comparison.

(a) shows intelligent task allocation results. The horizontal axis is the number of

allocated tasks, the vertical axis is device ID, and color depth indicates device workload.

Task allocation is uneven but reasonable, with the top 10 devices handling about 40% of

tasks. Devices with high computational capability and low energy - consumption get

more tasks, while those with high energy - consumption or low battery levels get fewer,

showing the algorithm's intelligence and fairness. (b) compares energy consumption

Lingli Yao

between traditional methods and the proposed model. Traditional methods allocate tasks

to the most powerful devices, leading to higher energy consumption. In contrast, the

proposed model saves about 31.2% of energy while maintaining overall system

performance by considering multiple factors.

5. Experimental Methods and Results Analysis

5.1 Performance Evaluation

As illustrated in Table 3, the comparative performance between our model and

traditional task allocation methods.

Table 3. Performance Comparison

Metric

Traditional Method

Proposed Model

Improvement

Energy consumption (J)

3245.6 ± 82.3

2232.7 ± 57.1

31.2%

Task completion time (s)

187.3 ± 12.6

195.1 ± 13.8

-4.2%

Workload balance index

0.62 ± 0.05

0.83 ± 0.04

33.9%

Network lifetime (h)

18.4 ± 1.2

26.7 ± 1.3

45.1%

Device utilization

0.72 ± 0.08

0.65 ± 0.05

Closer to optimal

The workload balance index measures the uniformity of task distribution across devices,

calculated as:

Balance Index=1− √1

n∑(

i=1 Li−L

‾)2

‾ （21）

where Li is the workload of device i, L

‾ is the average workload, and n is the total number

of devices. A value closer to 1 indicates better balance. The proposed model achieves 0.83,

demonstrating superior load distribution compared to traditional methods (0.76).

The energy saving rate demonstrates statistical significance (p < 0.001) and exhibits an

exponential relationship with system size. As shown in Figure 3, our model achieves

balanced task distribution while substantially reducing energy consumption compared to

traditional approaches that prioritize high-capability devices regardless of energy state.

An Adaptive Task Scheduling Based Energy Optimization Model for IoT Devices

5.2 Detailed Analysis of Task Completion Time Trade-off

Experimental results show task completion time increased from 187.3s to 195.1s (+4.2%)

compared to traditional methods. This slight increase is a reasonable trade-off for energy

optimization, with detailed analysis as follows:

Energy-Efficiency Priority Strategy:

The model prioritizes energy-efficient devices over purely high-speed devices, which is

reflected in the instantiated DFAEF-based efficiency metric:

ηi(t)=α⊆(1−Li(t))2⊆e−β⊆|Li(t)−ω|⊆(Ei

cur(t))2 （22）

Devices operating in the optimal efficiency range (Li≈0.65) are prioritized even if they

are not the fastest ones.

Load Balancing Consideration:

Traditional methods concentrate tasks on high-performance devices, quickly depleting

their batteries. Our model employs balanced allocation to avoid device overload:

S(i,j)=w1⊆Ci

comp +w2⊆Bi

data +w3⊆Ei

cur(t)+w4⊆(1−Li(t)) （23）

Long-term System Stability:

The 4.2% time increase trades for 31% energy reduction and 23.6% network lifetime

extension. From lifecycle perspective, this significantly improves overall system benefits.

IoT Application Scenario Adaptation:

Most IoT tasks (environmental monitoring, data collection) are not time-critical. The 4.2%

delay is acceptable, while energy savings are crucial for battery-powered devices.

Table 4.Performance Trade-off Quantitative Analysis:

Performance Metric

Traditional Method

Proposed Method

Improvement

Task completion time

187.3s

195.1s

+4.2%

Average energy consumption

245.7W

169.3W

-31.1%

Network lifetime

42.3h

52.3h

+23.6%

Load balance index

0.76

0.83

+9.2%

Lingli Yao

This trade-off aligns perfectly with IoT system design principles of "energy efficiency first,

sustainable operation." Therefore, the 4.2% increase in task completion time is an

acceptable overhead for achieving more than 30% energy savings and significantly

extended network lifetime in typical non-time-critical IoT scenarios.

6. Conclusion

This paper presents an adaptive task scheduling based energy optimization model for IoT

devices. By combining the Dual-Factor Adaptive Efficiency Function, a dynamic power

allocation algorithm, a multi-parameter weighted task scheduling algorithm, and network

topology optimization, the proposed model achieves a significant reduction in IoT system

energy consumption. Experimental results indicate that, compared to traditional

methods, the proposed model reduces energy consumption by approximately 31.2%

while maintaining good task allocation balance and network connectivity. At the same

time, the model accepts a slight 4.2% increase in task completion time as a deliberate

trade-off to obtain substantial gains in energy efficiency and network lifetime. The

findings of this research have important practical value for energy-constrained IoT

systems, effectively extending the lifespan of devices and networks. Future research will

further explore the application of reinforcement learning techniques in parameter

adaptive adjustment to adapt to more complex and variable network environments.

Simultaneously, considering more complex task dependency relationships and expanding

the applicability of existing scheduling algorithms also represent important directions.

Additionally, integrating energy harvesting technologies, designing more sustainable

energy management strategies, and conducting deployment tests of actual IoT systems to

verify model performance in real environments will be the focus of subsequent work.

These research directions will further advance the development of IoT energy

optimization and provide theoretical and technical support for constructing more

efficient and sustainable IoT systems.

Fund project:（1）Jiangsu Province Industry University Research Cooperation

Project-Research on the Method of Online Calibration Device for Surgical Shadowless

Lamp（BY20240184）

（2）2021 Excellent Science and Technology Innovation Team of Universities in Jiangsu

Province - Smart Granary Comprehensive Prevention and Control Information Perception

and Transmission Key Technology

（3）The second batch of provincial industrial and information industry transformation

and upgrading projects in Jiangsu Province in 2020 Intelligent Grain Big Data

Management Service Platform

References

[1] R. Kumar , D. Sharma, "Systematic review of IoT growth and future aspects," IEEE

Internet of Things Journal, vol. 9, no. 3, pp. 2050-2063, 2023.

An Adaptive Task Scheduling Based Energy Optimization Model for IoT Devices

[2] J. Wang, X. Li , H. Zhang, "Energy optimization in IoT networks: A survey," IEEE

Communications Surveys & Tutorials, vol. 24, no. 1, pp. 129-173, 2022.

[3] M. Chen , Y. Yang, "Battery-aware task scheduling in IoT systems: Challenges and

solutions," ACM Transactions on Sensor Networks, vol. 18, no. 2, pp. 1-25, 2022.

[4] Q. D. La, M. V. Ngo, T. Q. Dinh, T. Q. Quek,, H. Shin, "Enabling intelligence in fog

computing to achieve energy and latency reduction," Digital Communications and

Networks, vol. 5, no. 1, pp. 3-9, 2023.

[5] Z. Almudayni, B. Soh, H. Samra, A. Li, "Energy inefficiency in IoT networks: Causes,

impact, and a strategic framework for sustainable optimisation," Electronics, vol. 14,

no. 1, pp. 159, 2025.

[6] Z. Liu, Y. Chen, W. Zhang, "Heterogeneous multi-core architecture for low-power IoT

devices," IEEE Transactions on Circuits and Systems, vol. 70, no. 2, pp. 687-700,

2023.

[7] X. Wang et al., "ADAVS: Adaptive voltage scaling for IoT processors with dynamic

workload," IEEE Journal of Solid-State Circuits, vol. 57, no. 4, pp. 1046-1058, 2022.

[8] X. Zhang, J. Li, R. Qiu, T. S. Mean, F. Jin, "Optimized routing model of sensor nodes

in Internet of Things network," Sensors and Materials, vol. 32, no. 8, pp. 2801-2811,

2023.

[9] H. Pereira, G. L. Moritz, R. D. Souza, A. Munaretto, M. Fonseca, "Increased network

lifetime and load balancing based on network interface average power metric for

RPL," IEEE Access, vol. 8, pp. 48686-48696, 2023.

[10] S. Gupta, R. Garg, N. Gupta, W. S. Alnumay, U. Ghosh, P. K. Sharma, "Energy-

efficient dynamic homomorphic security scheme for fog computing in IoT networks,"

Journal of Information Security and Applications, vol. 58, article 102768, 2023.

[11] L. Zhang , D. Wu, "Task offloading strategies for battery-powered IoT devices,"

IEEE Transactions on Mobile Computing, vol. 21, no. 5, pp. 1763-1777, 2022.

[12] A. Rahman, M. Hossain, "Energy-efficient sensor network design for IoT

applications," IEEE Sensors Journal, vol. 22, no. 10, pp. 9572-9583, 2022.

[13] J. Chen, X. Li, B. Zhao, "E-AOMDV: An energy-aware multipath routing protocol

for IoT networks," Wireless Networks, vol. 28, no. 3, pp. 1231-1248, 2022.

[14] H. Kim, S. Park, "LP-MAC: A low-power MAC protocol for IoT networks," IEEE

Transactions on Wireless Communications, vol. 21, no. 6, pp. 4152-4165, 2022.

[15] K. Zhang, Y. Wang, J. Li, "Data aggregation approach for energy saving in IoT

sensors," IEEE Internet of Things Journal, vol. 9, no. 5, pp. 3623-3638, 2022.

[16] W. Li, S. Huang,L.Wang, "DeepEMS: Deep reinforcement learning based energy

management system for IoT networks," IEEE Transactions on Network Science and

Engineering, vol. 9, no. 1, pp. 165-178, 2022.

[17] R. Singh et al., "Adaptive computation offloading in battery-constrained IoT

systems," IEEE Transactions on Sustainable Computing, vol. 7, no. 2, pp. 314-327,

2022.

Lingli Yao

[18] K. Anderson, V. Kumar, A. Singh, "Energy-aware task scheduling for IoT devices

with machine learning," IEEE Transactions on Sustainable Computing, vol. 7, no. 3,

pp. 505-517, 2022.