Knowledge extracted from measured trajectories is categorized according to the support extent of system models. The importance of extracting information from disturbed trajectories without system models is emphasized. The idea of extracting knowledge after trajectory aggregation is summarized. Therefore， mode identification task of multi-machine trajectories in time domain is converted into that of single-machine trajectory with the stability-preserving dimensional-reduction transformation. Time-frequency analysis methods are adopted to extract time-varying dynamic characteristics from the image trajectory of the single-machine. The effects of trajectory aggregation and time-variation are studied. Credibility verification is needed when time-variation is strong. For verifying the model and parameter， proper index is needed to quantize the differences between two sets of trajectories. Then， sensitivity analysis of the index can be applied to identify the actual model and parameter.