Federated Intrusion Detection for Industrial IoT: A TCN-BiGRU Framework with Kalman Filter Optimization
Wentao Zhang, Guoqiang Xia, Peng Zhang, Jianhu Sun, Jindan Dong, Haoran Wang · 2025
This paper focuses on the intrusion detection problem in Industrial Internet of Things (IIoT) and SCADA networks, and proposes a TCN - BiGRU framework based on federated learning. A Kalman filter is introduced to optimize the performance of the model. This framework captures the long - term dependencies of time series data through the Temporal Convolutional Network (TCN), and combines with the Bidirectional Gated Recurrent Unit (BiGRU) to achieve bidirectional extraction of context features, which solves the limitations of traditional methods when dealing with high - dimensional dynamic network data. In the federated learning framework, the Kalman filter is applied to the dynamic optimization of the global model weights, aiming to improve the convergence speed and training stability, while ensuring data privacy and adapting to model heterogeneity. The experimental results show that the framework proposed in this paper has good performance in terms of detection accuracy and stability. This research provides a new technical path for privacy - preserving collaborative security protection in industrial control systems, and in the future, the application potential of this framework in cross - industry multi - modal attack detection will be further explored.