In order to reduce the impact of individual characteristic differences on the accuracy of the non-invasive blood pressure prediction model and improve the prediction accuracy, a Bayesian optimization (BO) XGBoost non-invasive blood pressure prediction method is proposed. Firstly, the multivariate linear model is established to obtain preliminary blood pressure prediction values based on pulse transit time (PTT) and body mass index (BMI). Then combine human characteristic parameters as the input of XGBoost blood pressure prediction model. Then use Bayesian optimization to automatically optimize XGBoost hyperparameters. Finally， the BO-XGBoost model is used to predict the blood pressure and compare with other methods. The experimental results show that the average absolute error of diastolic and systolic blood pressure based on the BO-XGBoost blood pressure prediction model meets the standard of less than 5mmHg formulated by AAMI (American Medical Instrument Promotion Association), which is better consistent with the method of mercury sphygmomanometer.