Construction of English Job Skills Training Platform and Data Security Mechanism for Textile Enterprises Based on Federated Learning

Distributed intelligent systems increasingly require collaborative learning across heterogeneous data sources while preserving data confidentiality, a challenge that is also relevant to networked sensing and communication environments. To address the security and efficiency limitations of centralized training in textile enterprises, this study develops a federated-learning-based English job skills training platform that integrates differential privacy and secure multi-party computation into a unified collaborative framework. Local model updates are protected through Gaussian perturbation, while segmented encryption and secret sharing secure parameter transmission and aggregation without exposing raw data. The proposed strategy enables stable global optimization while maintaining privacy guarantees under cross-enterprise collaboration. Experimental results show that the platform converges to a global loss of 0.1619 by the 44th training round and improves the average oral expression score from 73 to 87. Under a privacy budget of 0.50, the perturbation index is reduced to 0.39, outperforming representative federated baselines in privacy preservation while maintaining model utility. The integrated mechanism also demonstrates improved robustness against model poisoning attacks and lower aggregation overhead under multi-enterprise participation. Beyond enterprise training scenarios, the proposed privacy-preserving collaborative architecture provides methodological insights for secure distributed intelligence and data fusion in communication-oriented engineering systems where reliable information exchange and decentralized optimization are essential.