The support vector machine (SVM) is a small computational data set established under the classification problem, which can achieve nonlinear highlatitude classification with good scalability. However, during the training process of traditional SVM, the results of SVM computation are closely related to parameter selection, and the parameter selection algorithms currently in use have a number of defects. Aiming at above problems, we introduce the gray wolf algorithm (GWO) into the chaotic sequence, change the initial distribution of wolves, and propose a chaotic gray wolf optimization algorithm (CGWO), which can improve the uniformity of wolf distribution and the ergodicity of wolf searching, thus greatly enhancing the computing speed and accuracy of the GWO algorithm, and ultimately achieve better SVM optimization. Comparative experiments on the open source data provided by Mirjalili mixed with the original data as the test set of the vector machine shows that the CGWO algorithm has obvious performance improvement, and it outperforms the GWO algorithm,  artificial bee colony, gravitational search algorithm and SVM optimized by traditional optimization algorithms, with higher computation accuracy  lower error and less time.