In the context of the increasing complexity of cybersecurity threats and the surge in demand for data privacy protection, this paper proposes a privacy-preserving cybersecurity situation awareness model based on federated learning and designs an improved adaptive weighted federated learning algorithm (AWA-FL). The AWA-FL algorithm effectively solves the performance bottleneck of traditional federated learning under non-independent and identically distributed data by dynamically adjusting the client model weight update strategy. The experiment is based on the CICIDS2017 and UNSWNB15 datasets and simulates a distributed network environment. The results show that compared with the traditional federated average algorithm (FedAvg), the accuracy of this model is improved by 12.3%, the number of convergence rounds is reduced by 35%, and the success rate of data reconstruction attack is reduced by 40% under the condition of differential privacy budget ε=0.5. Compared with the centralized learning algorithm, the privacy leakage risk of this model is almost zero under the same data scale, and the high accuracy of 92.6% is maintained. In addition, the time complexity of the AWA-FL algorithm is reduced by 28% compared with the traditional algorithm, showing higher computational efficiency. The experiment fully verified that the model can significantly improve the accuracy and timeliness of network security situation awareness while ensuring data privacy.
Qi et al. (Fri,) studied this question.