Deep learning based intelligent intrusion detection (IID) methods have been received strongly attention for computer security protection in cybersecurity. All these learning models are trained at either a single user server or centralized server. For one thing, it is almost impossible to train a powerful deep learning model at a single user. For other, it will encounter intrusion risks at centre server and violate user privacy if collecting dataset from all of user servers. In order to solve these problems, this paper proposes an effective IID method based on federated learning (FL) aided long short-term memory (FL-LSTM) framework. First, the initial LSTM global model is deployed at all of user servers. Second, each user trains its single model and then uploads its model parameters to central server. Finally, the central server performs model parameters aggregation to form a new global model and distributes it to user servers. Use this step as a loop for communication to complete the training of the intrusion detection model. Simulation results show that our proposed method achieves a higher accuracy and better consistency than conventional methods.

基于深度学习的智能入侵检测（IID）方法已受到网络安全中计算机安全保护的强烈关注。所有这些学习模型都是在单个用户服务器或集中式服务器上进行训练的。一方面，几乎不可能在单个用户上训练强大的深度学习模型。另一方面，如果从所有用户服务器收集数据集，它将在中心服务器上遇到入侵风险并侵犯用户隐私。为了解决这些问题，本文提出了一种基于联合学习（FL）辅助长短期记忆（FL-LSTM）框架的有效IID方法。首先，初始的LSTM全局模型部署在所有用户服务器上。其次，每个用户训练其单个模型，然后将其模型参数上载到中央服务器。最后，中央服务器执行模型参数聚合以形成新的全局模型并将其分发给用户服务器。使用此步骤作为循环，以完成入侵检测模型的训练。仿真结果表明，与传统方法相比，本文提出的方法具有更高的精度和更好的一致性。