In order to meet the requirements of high-precision alignment and positioning of optical elements for two-mirror reflection system, as well as the requirements of reliability and efficiency for system engineering applications, an adaptive alignment technology based on artificial neural network (ANN) was proposed. Based on the vector wave aberration theory, the mapping relationship between the wave aberration and the offset of the two-mirror system was analyzed. The ANN was built under the framework of Keras, and the adaptive alignment model was constructed with non-analytic ideas. An adaptive alignment device was developed to make the alignment accuracy of the secondary mirror better than 2 μm, and the tilt alignment accuracy is better than 2″, which solved the technical problems such as algorithm design and precision optimization, micro-stress connection of mirror group, and completed the adaptive alignment verification of a double parabolic defocusing system. The test results show that by using this technology, the wave aberration after alignment of system is better than λ/16, the alignment cycle is greatly shortened, and the assembly reliability passes the environmental test assessment, which lays a foundation for the engineering application of the technology.