Annotation

Annotation: The aim of the research is to improve the quality of streaming data processing in intelligent systems based on federated distributed computing — a synthesis of fog computing principles and federated learning. A model for quality optimization is proposed, distinguished by the integration of performance, accuracy, and resource consumption criteria into a single integral loss functional. Methods and algorithms have been developed for adaptive federated streaming data processing, combined learning mode, streaming feature transformation, associative federated aggregation of statistical structures, and an incremental decision tree, which together ensure coordinated parameter adaptation, robustness to asynchrony and concept drift, reduced latency, and improved quality of streaming data processing. Experimental validation on a heterogeneous testbed demonstrated a reduction in average latency and an increase in throughput while maintaining stable accuracy and complying with resource constraints. Recommendations have been formulated for the development of standards ISO/IEC TS 8200:2024, GOST R 59277–2020, and GOST R 57700.27–2020. It has been established that coordinated adaptation of synchronization frequency, compression ratio, and forgetting coefficient, combined with streaming preprocessing and associative aggregation, ensures a stable decrease of the integral loss functional. Further research is aimed at refining the applicability boundaries of the proposed solutions across load profiles and topologies, as well as enhancing reliability and fault-tolerance mechanisms.