Abstract: This paper proposed a new method to diagnose the number of broken rotor bars of induction motors based on estimation of signal parameters via rotational invariance technique（ESPRIT）, pattern search algorithm（PSA） and light gradient boosting machine（LightGBM）. The performance of ESPRIT-PSA was tested with the simulated instantaneous reactive power signal under the broken rotor bar fault situation. The results show that ESPRIT-PSA can clearly identify the broken rotor bar fault associated component in the instantaneous reactive power signal even with short-term sample. Recently, motor instantaneous reactive power signal analysis（MIRPSA） has been used to detect the broken rotor bar fault of induction motors due to its effectiveness in low slip situations. However, this type of method cannot accurately diagnose the number of broken rotor bars. Therefore, LightGBM was introduced to classify the broken rotor bar fault so as to accurately diagnose the number of broken rotor bars. Finally, the experiment was performed to diagnose the broken rotor bar of an induction motor. The results show that the method is effective and suitable for low slip situations for it uses instantaneous reactive power as the analysis signal.