Abstract: When data is volatile, the specific set of regression parameters of the traditional support vector machine cannot meet the requirement for the parameters to change with the data distribution. This results in the regression curve not meeting the precision requirements. At the same time, we wanted to remove some of the nonessential data in the regression process to speed up the problem-solving process. To address the above two problems, in this paper, we present a vector pre-selected piecewise regression algorithm for a support vector machine (p-p-SVR). First, the algorithm deletes some unnecessary data, based on their spatial distribution. Next, based on the complexity of different regions of the sample, training data are divided into several domains, and corresponding parameters are set for each region. Simulation results show that, compared with the traditional method, the p-p-SVR algorithm has better regression accuracy and generalization performance.