The invention discloses a method for detecting multi-period oscillation behaviors of industrial control loops. The method includes the following steps: (1) collecting a set of process data xt in the to-be-detected control loops, (2) carrying out wavelet packet conversion decomposition on the process data xt to obtain an original binary-tree-shaped decomposition structure T0, (3) simplifying the original binary-tree-shaped decomposition structure T0 according to the minimum shannon entropy rule, the non-Gaussian indexes and the quasi oscillation indexes to obtain an optimal binary-tree-shaped decomposition structure Tbest, and (4) signing subspace corresponding to non-Gaussian distribution leaf nodes according to the non-Gaussian indexes TG of all leaf nodes on the optimal binary-tree-shaped decomposition structure Tbest to obtain a coordinate position set I corresponding to the leaf nodes with the signed subspace. By means of the method, the multi-period oscillation behaviors of the industrial control loops can be quantitatively detected, the regulation degrees and the periods of oscillation components of multi-period oscillation are obtained, and abundant data supports are provided for evaluation and fault source diagnoses of the oscillation behaviors.