Nowadays, network security has risen to the national security strategy level. As a significant part of network security, the intrusion detection technology has aroused general concern. Based on the research of the immune mechanism intrusion detection, aiming at the problems concerning the slow generation of the detector and the low detection efficiency caused by the traditional real-valued negation selection algorithm is not conducive to the efficient analysis of the data, this paper introduces the local linear embedding algorithm which can be applied to reduce the high dimensional data preprocessing optimization dimension due to the characteristic of map the dimensionality of high-dimensional data, and combines with the real-valued negative selection algorithm to generate the detectors. Then, using this algorithm to detection model can enhance the generation velocity of the detectors and ensure the generated detector to process the high-dimensional data efficiently. The algorithm can ensure the local linear structure of the sample is the same after the dimensionality reduction, and it also has the characteristics of less variable parameters and shorter computation time. The experimental results show that this algorithm can significantly improves the detector generation velocity and the detection efficiency of the data, and it is also outstanding in the detection performance.