Efficient stable case screening algorithm can reduce the computational burden of transient stability assessment if the actual unstable fault is not recognized as stable. The quantization accuracy can be improved while both the integral and mapping steps decrease in extended equal area criterion(EEAC)algorithm. However

when the integral and mapping steps are both decreased to a certain extent

the quantization accuracy may no longer be improved obviously. Comparing the results of static EEAC(SEEAC)

which neglects all time-varying factors

and dynamic EEAC(DEEAC)

which partially considers the time-varying factors

if both stability margins obtained by these two algorithms are large enough and their difference is small enough

the case can be confirmed as stable

and the complete process of EEAC algorithm can be early terminated. The sufficient but unnecessary rules for filtering stable cases can be acquired when this causal relationship is introduced to the machine learning system

and their related threshold values can be optimized with statistical analysis. Then the stable cases can be hierarchically screened out. Its robust performance with the unified rules and threshold values is verified by 28 sets of data from 7 actual Chinese provincial power systems under various operating conditions

that is

for cases with the two ends of the faulted branch opened at different time

no any actual unstable cases is filtered out and the whole computational burden is reduced by 72. 1% comparing to the complete process of EEAC algorithm.