Abstract: High-quality representation and precise characterization of the time-frequency (TF) information in seismic signals are crucial for seismic data processing and interpretation. Traditional TF analysis methods generally suffer from insufficient resolution, while synchrosqueezing theory has effectively mitigated this limitation. Notably, the multisynchrosqueezing transform achieves higher precision than conventional synchrosqueezing algorithms by iteratively applying the synchrosqueezing operation to the original TF spectrum. To this end, this paper proposes a multisynchrosqueezing transform algorithm based on the generalized S-transform, where an entropy criterion is introduced to adaptively optimize the time window. The proposed algorithm is applied to extract Teager-Kaiser (TK) dominant energy attributes from seismic data for reservoir prediction. Results from synthetic signal tests show that the proposed method not only exhibits excellent TF representation capability but also enables signal reconstruction. Even under noise contamination, it can still effectively characterize the TF distribution of the effective signals. A Marmousi2 model test demonstrates a strong correlation between reservoir properties and the TK dominant energy attributes derived using the proposed method. Application results from 2D field data further validate its effectiveness in reservoir characterization, thus providing reliable technical support for reservoir identification in hydrocarbon exploration.