In order to solve the problem that the radial velocity of the active phase of the rocket is sometimes missing in the continuous wave radar test, the coordinates of the trajectory measurement radar test jointly participated in the test are selected as the feature vector, and the radial velocity is selected as the target vector. The data of the two radars are fused, and the univariate linear regression model of velocity and range, velocity and transverse deviation, as well as the support vector regression model of velocity and range, transverse deviation are established by using sample 1, Then take sample 2 as the test data, take the predicted values of the three models as the feature vectors, and the corresponding measured values as the target vectors, and establish the genetic algorithm optimized LSSVM model. Finally, combine samples 1 and 2 as the training data, and sample 3 as the test data, and bring the predicted values of the two linear regression models and support vector regression machine models into the genetic algorithm optimized LSSVM model, The radial velocity of sample 3 predicted by LSSVM optimized by genetic algorithm is obtained. Finally, the predicted values of sample 3 are combined by the four models to obtain the joint predicted values of multiple models. The experimental results show that the accuracy of the joint prediction value of multiple models is the highest, with an error of 0.065%, less than 1 ‰, which meets the error requirements of continuous wave radar for measuring the radial velocity of rockets.