X射线自由电子激光XFEL(X-ray Free Electron Laser）是目前研究的热点。光阴极注入器是XFEL的首选电子源，其所提供的低发射度高品质电子束流对于XFEL非常重要，在光阴极注入器中有诸多参数影响着束流的发射度，为获得低发射度高品质束流需要进行参数优化。对于多参数的优化问题通常会采用算法来提升优化效率，比如常用的遗传算法，但是当参数较多时，其计算量仍旧很大。本项目提出如下方案用以进一步提升优化效率：利用机器学习来建立并训练束流动力学软件的替代模型，再把此模型用于遗传算法的优化过程中。这一思路尤其适合多目标优化，且具有较好的拓展性，未来可以进一步应用于加速器的控制、调谐等方面。

XFEL (X-ray Free Electron Laser) is nowadays a very hot research field, and photo-injector is the best choice for the electron source of the XFEL. Low emittance and high quality electron beam from the photo-injector is very essential for the XFEL, there are many parameters of the photo-injector which can influence the emittance of the electron beam, getting low emittance all these parameters should be optimized. For the multi-parameters optimization problems we usually use some algorithm, e.g. the Genetic Algorithm (GA), to promote the optimization efficiency, but when the optimization has lots of parameters to optimize the calculation is still very large. Here we propose a proposal to further improve the optimization efficiency: First, using machine learning to build and train a surrogate model of the beam dynamic simulation software, then the surrogate will be used in the GA optimization process. This proposal is very appropriate for multi-objective optimization, and it also has good compatibility, in the future it can be used in the control and tuning of the accelerator.