With the increasing diversity and rapid variability of network intrusion, the development of real-time network security monitoring systems with high flexibility and strong adaptability still faces severe challenges. Therefore adaptive network intrusion detection(ANID) method based on fuzzy rough set attribute reduction(FRS-AR) and Gaussian mixture model linear discriminant aualysis(GMM-LDA) optimal cluster feature learning(GMM-LDA-OCFL) is proposed. Based on the fuzzy rough set theory, the optimal attribute set of network connection data is selected automatically by information gain rate measurement. Then, an optimal cluster feature learning method based on GMM-LDA is proposed to obtain the optimal feature representation of the normal mode feature library and the intrusion mode feature library. At the same time, the adaptive on-line update mechanism of the normal(abnormal) pattern feature library is introduced, so that the detection model can adapt itself to dynamic network changes. The test results of KDD99 and network simulation experiment platform based on Nidsbench show that the proposed method can effectively adapt to the dynamic changes of the network environment and various intrusion behaviors in the real network connection data can be detected in real time. And the performance of the proposed method is better than that of the existing commonly-used intrusion detection methods, which has potentially wide application prospects.