IoT-Enabled Hierarchical Architecture for Intelligent Home-Based Elderly Care: A Multi-Objective Optimization Approach

Abstract

With the acceleration of global aging process, the contradiction between the home care needs of disabled elderly and limited nursing resources is becoming increasingly prominent. To address monitoring blind spots, delayed responses, and lack of personalized services in traditional elderly care models, a smart elderly care ecosystem based on the Internet of Things is constructed. By integrating data collection, transmission, and intelligent service modules through a layered architecture design, a multi-modal sensor network is deployed to fuse physiological parameters, behavioral trajectories, and environmental data. The research involves deploying Raspberry Pi 4B edge nodes in 15 households of disabled elderly individuals, integrating wearable devices, ultra-wideband positioning tags, and environmental sensors, totaling 23-28 units per household. The sampling frequency is 1 Hz during the day and 0.017 Hz at night, with an average data volume of 12.7 GB per day. The core algorithm includes an improved LZW compression algorithm that reduces data redundancy through differential preprocessing and dynamic dictionary elimination, a dynamic priority scheduling mechanism that uses a random forest classifier to identify event urgency and predicts pre-allocated bandwidth based on LSTM behavior, and a multi-objective particle swarm optimization algorithm for balancing energy consumption and load distribution. The proposed system was deployed in home-based elderly care for disabled individuals. The results showed that the improved compression algorithm and dynamic priority scheduling mechanism reduced the compression rate by 40.09% and shortened the transmission delay of key data in network jitter scenarios by 61.3% at a sampling frequency of 6 times/min. After introducing a multi-objective optimization load balancing strategy, the day and night energy consumption were reduced by 30.8% and 27.5%, respectively. In the 12-month controlled experiment, a single-group pre-post design was adopted, with 30 participants aged 72.5±6.8 years. Based on the MIT-BIH arrhythmia database and the UR fall dataset, the training set was constructed to verify that the prediction accuracy of chronic diseases increased to 81.5% (the original baseline was 67.7%, p<0.001), the incidence of bedsores decreased by 78.9% (the original baseline was 37%), and the nursing cost decreased by 62.1% (the original baseline was 9,354 RMB/month). The study proposes a technical approach to mitigate resource mismatches in home-based elderly care services by constructing a closed-loop management system of "monitoring-warning-intervention", promoting the intelligent elderly care towards ecological and precise direction.