It is well known that it is difficult to achieve disturbance-detecting and algorithm switching in autoregressive moving average（ARMA） method combined with Prony method. A new method is proposed to solve this difficulty. The method is based on normalization kurtosis to judge whether ringdown signal exists in the sliding window. With this method， a low frequency identification scheme is developed to achieve adaptive switching between the common ARMA method and the higher order ARMA method. Case studies are given to compare the performances of the proposed method with Prony method as well as normal ARMA method. The results demonstrate the validity of the proposed scheme.