With the fast development of national power industry,ultra high voltage power transmission has already been applied in practice. And the following features such as stability and security of power system have become more crucial owing to the complexity of the grid. On the one hand, voltage stability is the major factor accounting for the power system reliability. On the other hand, the online assessment of the voltage stability in real time has always been an obstacle in the concerning research. This paper aims to put forward a QPSO-SVM model which can be applied to the online assessment of the voltage stability in real time, based on the increasing accuracy and efficiency of calculation by means of SVM model as well as the production of parameter via the method of QPSO. In addition, it ensures the absolute assessment and the comprehensive application to all networks adopt the component of tangent vector of power flow power as VSI so as to improve the assessment accuracy of machine learning. Finally, it is approved that by means of WSCC9-bus, the learning time, the assessment time and the accuracy have been increased by 23.2%, 63%, 77.9%, and 26.2%, 56.9%, 72.56% and 28.9%, 42.19%, 82.34%, respectively, compared with GA-SVM, SVM and BPNN. Also, the method based on the IEEE14 experiment is an ideal path for the online assessment of the voltage stability for the power system in real time due to the fact that the key buses can be found before the system collapses and that it shares the same findings with the power flow calculation.