Optimizing Energy Efficiency in IoT Networks through Federated Learning and Edge

Computing: A Review
Gummarekula Sattibabu B. Sunil Pratap A Arunkumar Gudivada
Department of ECE Department of ECE Department of ECE
Aditya College of Engineering & Aditya College of Engineering & Aditya College of
Technology Technology Engineering & Technology
Surampalem, India. Surampalem, India. Surampalem, India.
[email protected] [email protected] [email protected]

Abstract:

The rapid proliferation of Internet of Things (IoT) devices has led to the emergence of complex, data -
intensive networks that demand innovative solutions to manage energy consumption efficiently. This
paper provides a comprehensive review of the state-of-the-art machine learning techniques employed
to enhance energy efficiency in IoT-enabled networks. We explore various approaches, including
supervised, unsupervised, and reinforcement learning, highlighting their applications in optimizing
communication protocols, resource allocation, and network management. Key advancements in
federated learning and edge computing are also discussed, emphasizing their roles in reducing the
energy overhead associated with data transmission and processing. By synthesizing recent resear ch
findings, we identify critical challenges and potential future directions for integrating advanced
machine learning algorithms into IoT ecosystems. This review aims to provide researchers and
practitioners with a consolidated understanding of current trends and innovations, fostering the
development of more sustainable and energy-efficient IoT networks.

Keywords - Energy efficiency, Edge-Computing, Federated Learning, Internet-of-Things.

1. INTRODUCTION
The Internet of Things (IoT) encompasses a vast network of interconnected devices, including
sensors, actuators, smart appliances, and embedded systems, facilitating seamless data exchange for
various tasks and services, revolutionizing industries with smart environments like homes, cities, and
industrial automation [1]. However, the rapid proliferation of IoT devices poses significant
challenges, particularly concerning energy consumption [2]. These devices are often resource-
constrained, with limited battery life and computational capabilities, necessitating a focus on energy
efficiency in critical sectors such as healthcare, agriculture, and smart cities [3].

Figure 1. The global IoT connections 2022-2033, with forecasts to 2030. And Architecture of IoT

The global count of Internet of Things (IoT) devices is expected to nearly double, rising from 15.9
billion in 2023 to over 32.1 billion by 2030. These devices are integral across various industry sectors
and consumer markets. In 2023, the consumer segment alone accounted for roughly 60% of all IoT
or connected devices [25].
Machine learning (ML) techniques, including supervised, unsupervised, and reinforcement learning,
are pivotal for achieving energy efficiency in IoT networks [4]. Among these, federated learning (FL)
stands out due to its decentralized approach, which minimizes data transmission and utilizes local
computational resources [5]. By training models on IoT devices and sending only updates, FL reduces
energy costs, enhances data privacy, and extends battery life, making it a sustainable solution for IoT
ecosystems [6].

the objectives of this review include:

 Introducing centralised machine learning methods for optimizing energy consumption in IoT
networks. And
 Introducing federated learning and edge computing to provide significant advantages over
centralized machine learning methods in IoT networks.

The subsequent sections of the document are structured as follows: Section 2 provides a concise
overview of the relevant literature pertaining to centralized machine learning techniques and the
corresponding research. Section 3 outlines the utilization of federated learning and edge computing
as means to offer substantial benefits in comparison to centralized machine learning approaches
within IoT networks. Section 4 explains potential future paths, while the final section, Section 5,
summarizes the key findings of the study and proposes potential areas for further investigation.

2. RELATED WORKS ON CENTRALISED MACHINE LEARNING METHODS

Centralized machine learning methods, including supervised, unsupervised, and reinforcement

learning, are crucial for optimizing energy consumption in IoT networks. These techniques predict
energy patterns, identify efficient communication protocols, and dynamically schedule tasks to
enhance network performance and efficiency.

2.1 Supervised machine learning

Supervised machine learning algorithms play a crucial role in predicting energy consumption patterns
in IoT networks. Various studies have highlighted the effectiveness of different algorithms in this
domain. For instance, research by Gonzalez-Palacio et al. emphasizes the use of multiple linear
regression, support vector regression, random forests, and artificial neural networks to calculate path
loss and shadow fading, aiding in setting transmission power efficiently [7]. Additionally, El Mahjoub
et al. showcase the superiority of Long Short-Term Memory (LSTM) neural networks over Gated
Recurrent Unit (GRU) and Drop-GRU models for time series power forecasting, ensuring better
precision and fewer prediction errors [8]. Furthermore, Malki et al. demonstrate the success of the
Prophet and Light GBM models in anomaly detection and future energy consumption prediction
based on IoT sensor data, enhancing energy management in smart cities [9].

2.2 Unsupervised machine learning

Unsupervised machine learning techniques can indeed play a crucial role in identifying energy-
efficient communication protocols in IoT networks. Various studies have proposed novel methods
utilizing unsupervised learning for fault prediction [10], anomaly detection [11], and energy
management [12] in IoT applications. Additionally, the use of explainable feature selection and
adaptive model compression further contributes to the trustworthiness and accuracy of energy-
efficient communication protocols in IoT networks [11] [12].

2.3 Reinforcement learning

Reinforcement learning plays a crucial role in optimizing energy consumption and enhancing
network performance in IoT networks. By leveraging reinforcement learning algorithms, such as
Deep Q-Learning [13], fuzzy-based reinforcement learning [14], and Q-learning enabled intelligent
energy attack [15], IoT networks can dynamically schedule tasks, allocate resources efficiently, and
improve routing strategies. These approaches address challenges like energy efficiency, low-latency
task scheduling, and maximizing network lifetime.

2.4 Drawbacks of Centralized Machine Learning Methods

Centralized machine learning methods suffer from several drawbacks that hinder their efficiency and
applicability in IoT networks. These include high energy consumption due to continuous data
transmission and centralized processing, which require substantial computational power and cooling
systems. This setup also introduces latency issues, as data transmitted over long distances can delay
real-time decision-making. Additionally, centralized methods face bandwidth constraints when
handling large data volumes, leading to network bottlenecks and reduced performance. Moreover,
data centralization raises significant privacy and security concerns, exposing sensitive information to
potential breaches and misuse.

3. FEDERATED LEARNING AND EDGE COMPUTING

This section explores these two concepts in detail, discussing their roles and benefits in IoT-enabled
networks.

3.1 Federated Learning

Federated learning is a distributed method in which models are trained on various devices without
the necessity to transfer data to a central server [16]. Every device trains a local model using its own
data and then sends updates to a central coordinator, which combines them to enhance the overall
model. FL enhances energy efficiency in IoT networks by minimizing data transmission, lowering
communication energy costs, and keeping sensitive data local for enhanced security. It reduces the
need for extensive data transfers by transmitting only model updates, thereby saving energy and
optimizing communication load. FL also leverages edge devices' computational power, distributing
the workload and reducing energy consumption on central servers, promoting decentralized learning
in sensitive IoT environments.

3.2 Edge Computing

Edge computing is all about processing data closer to where it's generated, at the edge of the network,
instead of relying on centralized cloud servers. This helps to reduce delays, save on bandwidth, and
improve the overall efficiency of IoT networks [17]. By processing data locally, edge computing
minimizes the need for long-distance data transmission, reducing the energy consumption associated
with centralized cloud processing. Local data processing by edge devices lowers the volume of data
sent to central servers, decreasing energy costs [18]. It also enables real-time decision-making,
essential for applications like autonomous vehicles and industrial automation. Additionally, by
filtering and aggregating data at the edge, only essential information is transmitted to central servers,
optimizing bandwidth and conserving energy [19].

3.3 Synergy between Federated Learning and Edge Computing

Federated learning (FL) and edge computing can optimize energy consumption in IoT networks by
implementing innovative techniques. FL leverages distributed computing to preserve user privacy in
IoT networks [20] and [24], while edge computing enables real-time AI services by training models
on IoT data [21] and [23]. Techniques like “simultaneous wireless information and power transfer”
(SWIPT) in FL can enhance energy efficiency by charging devices during model training [20].
Furthermore, the implementation of energy harvesting techniques on IoT devices plays a crucial role
in promoting sustainable learning. Advanced algorithms such as joint device selection and resource
allocation are utilized to enhance energy efficiency without compromising the overall learning
experience. [22]. By combining FL's privacy-preserving capabilities with edge computing's real-time
processing, IoT networks can achieve significant energy savings and improved learning efficiency.

3.4 Overcoming Drawbacks with Federated Learning and Edge Computing

Federated learning (FL) and edge computing provide significant advantages over centralized machine
learning methods in IoT networks. FL allows models to be trained locally on devices, transmitting
only model updates to a central coordinator, thereby minimizing data transmission and reducing
associated energy consumption. Similarly, edge computing processes data close to its source,
reducing the need for long-distance data transfers and conserving energy while improving latency for
real-time applications. Both FL and edge computing distribute processing tasks, leveraging local
computational resources to reduce the load on central servers and optimize energy use. Moreover,
these approaches enhance privacy by keeping sensitive data on local devices and minimizing the risks
associated with central data storage, thereby offering more secure and efficient solutions for IoT
applications.
Table: 1: Summary of Energy-Aware Federated Learning for IoT-Enabled Edge Networks

This table provides a concise overview of methods, contributions, challenges, and applications in
recent studies on energy-aware federated learning for IoT and edge networks.

Authors and
Methods Used Contributions Challenges Applications
Ref No.

SWIPT for energy SWIPT-based AirFL Signal

transfer and model for energy-efficient misalignment,
Xinran, Z., et
distribution; deep federated learning; channel fading, AirFL; SWIPT
al. [20]
reinforcement learning optimized resource random noise;
for resource allocation allocation transmission errors

Large data transfers

Data deepening and JD2P architecture for from energy- Real-time AI
Sujin, K., et prefetching for energy-efficient edge limited IoT training; energy-
al. [21] classification learning; reduced devices; efficient edge
efficiency energy consumption maintaining learning
learning accuracy

Sustainable lifelong
Sustainable
Energy harvesting; learning on IoT Radio resource,
lifelong
Li, Z., et al. efficient device devices; improved computing power,
learning; data
[22] selection and resource convergence rate and battery life
utility
allocation reduced training constraints
maximization
latency

Energy-conscious
federated learning Energy-efficient DNN Energy-efficient
Bandwidth and
Li, Y., et al. with D2D model training; D2D- DNN training;
energy capacity
[23] communication; near- assisted federated D2D federated
constraints
optimal learning learning learning
algorithm
Authors and
Methods Used Contributions Challenges Applications
Ref No.

Cost-effective Privacy leakage in

Optimized device
federated learning; central ML IIoT-tailored
Ji, X., et al. selection; bandwidth
reduced costs while methods; resource FL; cost-
[24] allocation for IIoT
respecting energy constraints in IIoT efficient training
systems
constraints networks

The table highlights diverse methodologies, contributions, challenges, and targeted applications
across studies focusing on energy-aware federated learning in IoT-enabled edge networks. Each study
addresses specific aspects crucial for enhancing efficiency, sustainability, and performance in
federated learning environments tailored for IoT and edge computing contexts.

4. FUTURE DIRECTIONS

Achieving a balance between energy efficiency and accuracy will necessitate the exploration of
hybrid learning approaches and the fine-tuning of resource allocation strategies under stringent
resource limitations. Critical to efficient data management will be the optimization of communication
protocols and bandwidth usage. Furthermore, efforts to accelerate convergence rates and minimize
training latency through innovative algorithms and distributed computing methods aim to
significantly boost the scalability and effectiveness of federated learning across diverse edge
computing environments.

5. CONCLUSION
This paper reviews cutting-edge machine learning techniques aimed at enhancing energy efficiency
in IoT networks. It explores supervised, unsupervised, and reinforcement learning applications in
optimizing communication protocols, resource allocation, and network management. Federated
learning and edge computing are highlighted as pivotal for reducing energy overheads in IoT data
transmission and processing. Future research directions include refining hybrid learning approaches,
optimizing communication protocols, and leveraging distributed computing to enhance IoT network
sustainability and efficiency. Addressing these challenges and exploring new innovations will foster
smarter, more energy-efficient IoT environments, with potential for integrating AI-driven
optimizations with edge computing architectures.

REFERENCES
[1]. Goel, S. S., Goel, A., Kumar, M., & Moltó, G. (2021). A review of Internet of Things: qualifying technologies and
boundless horizon. Journal of Reliable Intelligent Environments, 7(1), 23-33.
[2]. Albreem, M. A., Sheikh, A. M., Alsharif, M. H., Jusoh, M., & Yasin, M. N. M. (2021). Green Internet of Things
(GIoT): Applications, practices, awareness, and challenges. IEEE Access, 9, 38833-38858.
[3]. Qaim, W. B., Ometov, A., Molinaro, A., Lener, I., Campolo, C., Lohan, E. S., & Nurmi, J. (2020). Towards energy
efficiency in the internet of wearable things: A systematic review. IEEE Access, 8, 175412-175435.
[4] Mahmood, M. R., Matin, M. A., Sarigiannidis, P., & Goudos, S. K. (2022). A comprehensive review on artificial
intelligence/machine learning algorithms for empowering the future IoT toward 6G era. IEEE Access, 10, 87535-87562.

[5] Li, Y., Chen, C., Liu, N., Huang, H., Zheng, Z., & Yan, Q. (2020). A blockchain-based decentralized federated learning
framework with committee consensus. IEEE Network, 35(1), 234-241.

[6] Shirvani, G., & Ghasemshirazi, S. (2024). Powering the Future of IoT: Federated Learning for Optimized Power
Consumption and Enhanced Privacy. arXiv preprint arXiv:2405.03065.
[7]. Mauricio, Gonzalez-Palacio., Diana, Tobón-Vallejo., Lina, María, Sepúlveda-Cano., Santiago, Rúa., Long, Bao, Le.
(2023). Machine-Learning-Based Combined Path Loss and Shadowing Model in LoRaWAN for Energy Efficiency
Enhancement. IEEE Internet of Things Journal, doi: 10.1109/JIOT.2023.3239827

[8]. Sameh, El, Mahjoub., Larbi, Chrifi-Alaoui., Bruno, Marhic., Laurent, Delahoche. (2022). Predicting Energy
Consumption Using LSTM, Multi-Layer GRU and Drop-GRU Neural Networks. Sensors, doi: 10.3390/s22114062
[9]. Amer, Malki., El-Sayed, Atlam., Ibrahim, Gad. (2022). Machine learning approach of detecting anomalies and
forecasting time-series of IoT devices. alexandria engineering journal, doi: 10.1016/j.aej.2022.02.038
[10]. Neha, Sharma., Udit, Agarwal., Om, Jee, Pandey. (2023). Energy-Efficient and QoS-Aware Data Routing in Node
Fault Prediction Based IoT Networks. IEEE Transactions on Network and Service Management, doi:
10.1109/tnsm.2023.3268676

[11]. Zijie, Huang., Yulei, Wu., Niccolò, Tempini., Hui, Lin., Hao, Yin. (2022). An Energy-efficient And Trustworthy
Unsupervised Anomaly Detection Framework (EATU) for IIoT. ACM Transactions on Sensor Networks, doi:
10.1145/3543855
[12]. Muhammad, Zawish., Nouman, Ashraf., Rafay, Iqbal, Ansari., Steven, Davy. (2023). Energy-Aware AI-Driven
Framework for Edge-Computing-Based IoT Applications. IEEE Internet of Things Journal, doi:
10.1109/JIOT.2022.3219202
[13]. Bassem, Sellami. (2022). Energy-aware task scheduling and offloading using deep reinforcement learning in SDN-
enabled IoT network. Computer networks, doi: 10.1016/j.comnet.2022.108957
[14]. Mekala, Ratna, Raju., Sai, Krishna, Mothku. (2023). Delay and energy aware task scheduling mechanism for fog-
enabled IoT applications: A reinforcement learning approach. Computer networks, doi: 10.1016/j.comnet.2023.109603
[15]. Long, Li., Yung-Jeng, Luo., Jing, Yang., Lina, Pu. (2022). Reinforcement Learning Enabled Intelligent Energy
Attack in Green IoT Networks. IEEE Transactions on Information Forensics and Security, doi:
10.1109/tifs.2022.3149148
[16] Nguyen, D. C., Ding, M., Pathirana, P. N., Seneviratne, A., Li, J., & Poor, H. V. (2021). Federated learning for
internet of things: A comprehensive survey. IEEE Communications Surveys & Tutorials, 23(3), 1622-1658.

[17] Yu, W., Liang, F., He, X., Hatcher, W. G., Lu, C., Lin, J., & Yang, X. (2017). A survey on the edge computing for the
Internet of Things. IEEE access, 6, 6900-6919.
[18] Wu, Y. (2020). Cloud-edge orchestration for the Internet of Things: Architecture and AI-powered data
processing. IEEE Internet of Things Journal, 8(16), 12792-12805.

[19] Bilal, K., Khalid, O., Erbad, A., & Khan, S. U. (2018). Potentials, trends, and prospects in edge technologies: Fog,
cloudlet, mobile edge, and micro data centers. Computer Networks, 130, 94-120.

[20]. Xinran, Zhang., Hui, Tian., Wanli, Ni., Zhaohui, Yang., Mengying, Sun. (2024). Deep Reinforcement Learning for
Energy Efficiency Maximization in SWIPT-Based Over-the-Air Federated Learning. IEEE transactions on green
communications and networking, doi: 10.1109/tgcn.2023.3307428
[21]. Sujin, Kook., Won-Yong, Shin., Seong-Lyun, Kim., Seung-Woo, Ko. (2024). Energy-Efficient Edge Learning via
Joint Data Deepening-and-Prefetching. IEEE Transactions on Wireless Communications, doi:
10.1109/twc.2024.3367352
[22]. Li, Zeng., Dingzhu, Wen., Guangxu, Zhu., Changsheng, You., Qimei, Chen., Yuanming, Shi. (2024). Federated
Learning With Energy Harvesting Devices. IEEE transactions on green communications and networking, doi:
10.1109/tgcn.2023.3310569

[23]. Yuchen, Li., Weifa, Liang., Jing, Li., Xiuzhen, Cheng., Dongxiao, Yu., Albert, Y., Zomaya., Song, Guo. (2023).
Energy-Aware, Device-to-Device Assisted Federated Learning in Edge Computing. IEEE Transactions on Parallel and
Distributed Systems, doi: 10.1109/TPDS.2023.3277423
[24]. Xiuzhao, Ji., Jie, Tian., Haixia, Zhang., Dalei, Wu., Tiantian, Li. (2023). Joint Device Selection and Bandwidth
Allocation for Cost-Efficient Federated Learning in Industrial Internet of Things. IEEE Internet of Things Journal, doi:
10.1109/JIOT.2022.3233595

[25]. Statista. (2024, June 12). Number of IoT connections worldwide 2022-2033, with forecasts to 2030.
https://www.statista.com/statistics/1183457/iot-connected-devices-worldwide/