Integrating Digital Twin and Hypergraph Neural Network for Sustainable IoT Service Placement in Edge Computing: A Multi-Objective Approach Toward Human Ecology

This study aims to design and evaluate two novel frameworks for sustainable IoT service placement in edge-cloud-IoT ecosystems, both focusing on reducing energy consumption, carbon emissions, and environmental costs. This is an applied-developmental study with a quantitative approach. The first framework, named EcoTwin-Deploy, is a multi-objective service placement architecture assisted by Digital Twin technology, integrated with the AKG-NSGAIII optimization engine to enable proactive and sustainable service placement. By leveraging future condition prediction, this framework minimizes latency, energy consumption, deployment cost, and carbon emissions while maximizing reliability. The second framework, named HyperGraphDeploy, is a multi-objective optimization framework based on Hypergraph Neural Networks (HGNN) that employs hypergraph structures to represent complex many-to-many relationships among IoT devices, edge nodes, cloud resources, and service chains. By integrating structural representation learning with the HG-MaOEA evolutionary optimizer, this framework simultaneously optimizes seven objectives including latency, energy, cost, reliability, carbon footprint, load balancing, and migration overhead. Comprehensive experimental evaluation was conducted in MATLAB R2024a simulation environment, with all competing algorithms executed under identical system configurations, similar workload distributions, and equal network assumptions.Results showed that EcoTwin-Deploy achieved improvements of 13.64% in energy consumption, 16.17% in service latency, 19.27% in load balancing, 20.71% in execution time, and 28.30% in convergence time compared to the QIAVOA-SP baseline. The Friedman statistical test with a statistic of 97.42 and a p-value less than 0.0001 confirmed the statistically significant superiority of this framework with an average rank of 1.38 among 9 competing algorithms. Module removal study showed that the Digital Twin prediction module has the largest contribution to performance improvement, as its removal leads to an 11.86% increase in latency and a 13.75% increase in load balancing. Furthermore, HyperGraphDeploy achieved improvements of 1.33% in connectivity, 0.91% in coverage, 20.97% in adaptation time, 19.04% in movement cost, 10.2% in latency, and 33.3% in SLA violation rate compared to D-POA. In a 1000-node scenario, this framework reduced execution time by 18.7% and memory consumption by 26.0%. The Wilcoxon statistical test confirmed the statistically significant superiority of HyperGraphDeploy at a 95% confidence level. The results demonstrate that integrating predictive and structural intelligence with multi-objective optimization can improve IoT service placement quality in complex edge-cloud-IoT ecosystems and contribute to environmental sustainability in urban environments. From a human ecology perspective, these frameworks are not merely technical achievements but mediators in the human-environment relationship that can help reduce the ecological footprint of technology and promote urban health. The findings can be useful for urban planners, environmental organizations, and researchers in the field of human ecology.