Network security is an important issue in the current information society. The development of artificial intelligence benefits from the constantly increasing amount of data, the improvement of computing power, and the continuous innovation of algorithms. Network security situational awareness can monitor, identify, and analyze threats, attacks, and abnormal behaviors in real-time on the network. The existing perception systems rely on feature engineering and cannot effectively cope with complex and ever-changing forms of network attacks. The high degree of imbalance in network security data and severe noise interference. The methods of cyberattacks are constantly evolving, and previous defense methods are difficult to adapt to new crises at any time. Therefore, this article proposes to construct a perception model, combine artificial intelligence technology, optimize the data preprocessing process, continuously update and iterate, and optimize the network security situation awareness ability. This article mainly uses experimental testing and simulation training to compare and analyze the functions of perception models and the performance of algorithms. The experimental results show that the deep neural network model achieves an accuracy of 96.5% in malware detection. Overall, deep neural networks can perform the best in intrusion detection, malware detection, and spam filtering, and RBF-SVM networks can also have the most advantages in anomaly detection.