In recent years, according to the data released by the national Internet Information Center, cyber attacks have become an upward trend, which has brought great potential threats to users, and even caused incalculable losses to many users. In order that users can use the network more safely and avoid users from network attacks, the research of detecting network attacks has become very important and urgent. Aiming at the problem that the prediction accuracy of existing network attack detection methods is not high, the intelligent algorithm of public network attack data mining based on deep neural network is studied. The openness and self-organization of public network make the network vulnerable to virus interference and intrusion attacks. Accurate and efficient mining of attack data can ensure network security; Traditional methods use time-frequency directional beam feature clustering method to realize attack data mining. When the signal-to-noise ratio is low, the probability of accurate attack data mining is low. The existing network attack detection methods include static detection and dynamic detection, but both of them have some shortcomings, rely too much on rules, and have the problem of high false positive rate. Aiming at the shortcomings of traditional network attack detection, this paper introduces deep neural network technology into the field of public network attack detection. According to the characteristics of data mining technology and attack detection, this paper applies data mining technology to public network attack detection system. Make the system have the ability of self-learning and predict the impact of the overall attack scheme on the attacked public network.