Noise-immune task offloading in industrial PONs: particle-Kalman cascaded filtering with Nash equilibrium-driven cooldown

Industrial passive optical networks (PONs) are pivotal for enabling high-speed, low-interference communication in next-generation industrial networks. To meet time-critical requirements and suppress data noises from complex industrial environments, we propose a data-noise-immunized computation offloading strategy integrating a particle-Kalman cascaded filter (PKCF) and a game-theoretic collaborative offloading with cooldown (GTCOC). The PKCF uniquely combines particle filtering (for non-Gaussian noise) and Kalman filtering (for Gaussian noise), achieving 99.9% noise suppression accuracy (error rates of ±0.0003%). Concurrently, GTCOC dynamically balances task distribution via Nash equilibrium adjustments, reducing average delay by 33.4% and improving load balancing efficiency by 40% in high-density scenarios compared with those of Deep Q-Network (DQN) / heuristic baseline methods. Extensive simulations validate that the proposed approach outperforms existing solutions in robustness and scalability under multi-source industrial noises (e.g., mechanical vibrations and electromagnetic interference), offering a practical framework for reliable edge-cloud coordination in the industrial PONs.