Model-based Reinforcement Learning in FedMA Approach for Industry 4.0
L. Natrayan
Abstract
Industry 4.0 is the collaborative work of different sectors of supporting elements in the industries. As Industry 4.0 entirely depends on data sharing and data-driven approaches, it requires a decentralized system to handle data from different sectors in Industry 4.0. Thus, the federated learning-based approach is used in Industry 4.0, which opts for a decentralized manner by preserving sensitive data within the sectors themselves. The federated system lacks the adaptive working conditions necessary to improve its performance based on the environment in which it operates. Thus, the proposed work utilizes model-based Reinforcement Learning in the Federated learning system of Industry 4.0. The proposed work, based on the FedMA approach, acts effectively based on the action and reward from the global model. Thus, the proposed work showed 94.52% accuracy with 94.05% F1 score that reduced the FPR and FNR values with the reconstruction loss of 0.041