Hybrid Fuzzy-Bayesian Demand Modeling and Simulation-Driven Policy Co-Optimization for Dynamic IoT Data Streams

Abstract

Dynamic IoT data streams evolve under changing communication, computation, energy, reliability, and service-demand conditions. Existing methods usually optimize stream learning, offloading, or resource allocation separately, without explicitly inferring latent service demands. This paper proposes DFB-PCO, a demand-aware fuzzy-Bayesian policy co-optimization method for IoT data streams. It extracts online stream features, uses fuzzy membership and rule aggregation to represent ambiguous demand boundaries, and applies Bayesian posterior updating to capture temporal demand uncertainty. The hybrid demand posterior and uncertainty score then guide coordinated filtering, compression, task offloading, routing, and resource allocation. A policy-dependency structure, coordination regularizer, and primal-dual constrained optimization mechanism are introduced to reduce conflicting actions and control delay, energy, reliability, privacy, and resource constraints. Experiments on four IoT/IIoT datasets show that DFB-PCO outperforms representative reinforcement-learning baselines in system cost, latency, energy consumption, constraint satisfaction, and demand recognition.