Support vector machine (SVM) was used to study the stability problem of high pressure unloading diaphragm pressure reducing regulator in the turbine gas seal pressure reducing test system. The research mainly focused on multi-structural parameters variation

which was difficult to solve by the classical methods. By comparing SVM and back propagation (BP) neural network model in the scarce and incomplete data set

several conclusions were proposed: the prediction models of SVM can get 25.5% less error in average and are superior to BP neural network model in the situation of multi-structural parameters variation; the generalization of SVM is also better than BP's in sparse samples space; suggestions of adjusting parameters are given for achieving good stability

and better decision-making is provided by SVM models in design of pressure reducing regulator

with double parameter combinations and tri-parameter combinations.