The selection of data analysis length is the key to extracting structural vibration characteristic information， and the artificial selection of the data length for signal analysis will cause certain errors in the calculation results. To reduce the influence of subjective factors， an improved multi-scale permutation entropy （IMPE） method is proposed to analyze the length of vibration data. IMPE method has a strong advantage in dealing with nonlinear and non-stationary signals， so the vibration measurement data of the measured structure can be obtained by placing sensors. For the obtained vibration signal， the one-dimensional time series data is multi-scaled and then coarse-grained to determine its phase space reconstruction parameters. The vibration data with different lengths is selected and the multi-scale permutation entropy （MPE） entropy value is calculated respectively. It is found that the entropy value is sensitive to the change of the data length， changes with the increase of the data length， and finally it is going to stabilize. Here， the stable value is defined as a standard entropy value， and the entropy value satisfying standard entropy value of 97% is used as the effective entropy value. The entropy value that meets the accuracy requirements is selected， and its corresponding shortest data length is defined as the optimal data analytical length of the vibration data. This method is applied to the selection of the optimal data length of the simulation signal and the vibration signal of discharge engineering， and the accurate data analysis length can be selected for structure monitoring， and it has good universality.