Гібридна архітектура Edge-cloud для обробки часових рядів у телемедичних IoT-пристроях

Abstract

This paper proposes a hybrid edge-cloud architecture for processing time series data in telemedicine IoT systems based on wearable devices. The relevance of the study is driven by the need to ensure efficient, low-latency, energy-efficient, and scalable analysis of large volumes of biomedical data generated in real time. The proposed architecture enables the distribution of computational tasks between the edge layer and cloud infrastructure, taking into account signal characteristics. At the edge level, preliminary data processing is performed, including signal filtering using a Kalman filter, dimensionality reduction via Piecewise Aggregate Approximation (PAA), and anomaly detection using the Isolation Forest algorithm. In addition, an adaptive data transmission policy is implemented, based on signal variability estimation, which allows dynamic selection between transmitting full data segments or aggregated features to the cloud. At the cloud level, in-depth analysis of temporal dependencies is carried out using a hybrid CNN-LSTM model that combines the advantages of convolutional and recurrent neural networks for effective extraction of complex patterns in the data. Simulation results on publicly available datasets demonstrate that the proposed approach reduces processing latency to 120 ms, decreases network traffic by approximately 65 %, and achieves anomaly detection accuracy of 91 %. The obtained results confirm the effectiveness of adaptive workload distribution between edge and cloud layers, ensuring a balance between system responsiveness, analytical accuracy, and energy efficiency. The proposed architecture can be applied in modern telemedicine monitoring systems, particularly under conditions of limited network resources and strict real-time processing requirements.