A Hybrid PSO-BP Optimized Fuzzy Neural Network for Network Security Situation Awareness

Abstract

With the continuous maturity of computer technology, computers are constantly subjected to network attacks. To better respond to different types of computer network security attacks, a computer network situational awareness model based on particle swarm optimization-error back propagation algorithm is proposed. These two algorithms are optimized by introducing compression factors and momentum. Meanwhile, the model adopts a hybrid update strategy, randomly selecting the PSO or BP algorithm for parameter update at each step. In the experimental results, when the number of iterations of the proposed PSO-BP model exceeded 30, the loss function value, accuracy, recall, and F1-score converged to around 0.01, 0.99, 0.94, and 0.98, respectively. Each performance index is superior to the other two PSO-BP algorithms. In addition, compared with other models such as adaptive fuzzy system, reinforcement learning-based method, CNN, LSTM, and Transformer, the PSO-BP model demonstrates higher detection accuracy and adaptability when dealing with dynamic network attack data. The proposed algorithm has demonstrated superior perception ability in the computer network security situation and against different types of network attacks. In practical situations, the research can provide timely and accurate situational awareness information for network administrators, helping them make quick decisions and reduce losses caused by network attacks. Meanwhile, the real-time performance of the model needs to be further optimized to better adapt to rapidly changing network security threats.