Adaptive Energy-Aware Federated Learning Framework for Traffic Flow Prediction in Smart Cities

Thant, Shwe Moe (2025) Adaptive Energy-Aware Federated Learning Framework for Traffic Flow Prediction in Smart Cities. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (1MB) | Preview
Preview	PDF (Configuration Manual) Download (1MB) | Preview

Abstract

Traffic flow prediction increasingly adopts federated learning because it enables distributed model training without centralising sensitive mobility data. However, existing approaches still face challenges in communication overhead, device heterogeneity and limited energy awareness. Prior studies often focus on improving modelling accuracy through graph-based or meta-learning methods, or enhancing privacy through mechanisms such as differential privacy and encryption. Yet, they rarely examine variations in client energy consumption or adapt participation to real device conditions.

This study introduces the Adaptive Energy-Aware Federated Learning framework, which integrates a lightweight GRU predictor with client-side energy estimation, bandwidth-aware metadata and selective client participation. Differential privacy is incorporated to assess how privacy noise affects model performance and energy usage. The system uses containerised IoT clients and an S3-based coordination layer, and is evaluated on the Los-Loop, PeMSD8 and SZ-Taxi datasets against two baselines, FedAvg and FedProx.

Results show that AEFL maintains accuracy comparable to the baselines while reducing energy consumption in datasets with substantial client workload variation. Its behaviour converges to baseline performance when workloads are uniform. Differential privacy introduces modest changes in accuracy and energy at moderate noise levels. These findings demonstrate that adaptive and energy-conscious federated learning provides a practical and sustainable approach to traffic flow prediction in smart-city environments.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Kazmi, Aqeel UNSPECIFIED
Subjects:	T Technology > T Technology (General) > Information Technology > Cloud computing H Social Sciences > HC Economic History and Conditions > Natural resources > Power resources > Energy consumption H Social Sciences > HE Transportation and Communications > Urban Transportation
Divisions:	School of Computing > Master of Science in Cloud Computing
Depositing User:	Ciara O'Brien
Date Deposited:	01 Sep 2026 11:40
Last Modified:	01 Sep 2026 11:40
URI:	https://norma.ncirl.ie/id/eprint/9742

Actions (login required)

View Item