The common detection method of sub-synchronous/super-synchronous oscillation is that the actually measured data is filtered firstly to obtain the sub/super-synchronous oscillation signal component.Then

the occurrence of the sub/super-synchronous oscillation is defined in accordance with the parameter identification result of each component signal. Thus

the accuracy of oscillation judgment is affected by both the effect of filter and the accuracy of parameter identification method.For solving above problem

a deep learning method

probabilistic neural network（PNN）combined with SST filtering is proposed to monitor the sub/super-synchronous oscillations of power systems. First

SST transform is used to replace traditional filter to achieve filtering function and the signal component of sub-synchronous frequency

power frequency and super-synchronous frequency are obtained. Then

parameter identification is not performed anymore

but the obtained each type of signal component is submitted to PNN neural network directly for classification

and the characteristics of sub/super-synchronous oscillation data are directly mined to achieve the purpose of online monitoring. The reliability of the method is verified by simulation and example.