Abstract: In this day and age, the security of Internet is faced with a growing number of threats from various sources. As a newly-emerging Internet security device which can dynamically monitor and block abnormal Internet transmission operations.Intrusion Detection System (IDS) has drawn enormous attention from the researchers. At present, the mainstream of IDS detects and handles the intrusion by matching the rules in the pre-made rules set. IDS based on such methods are low-efifcient, and are dififcult to maintain and strike a balance between speed and cost. To combat prolblems above, we propose a brand-new intrusion detection method which integrates the high-speed data stream capturing and ifltering based on hardware, classical machine learning techniques and Auto-encoder algorithm in Deep Learning, and realize a intelligent and high-speed IDS. Our syetem yield a good result which outperforms other IDS on the same cost level.