Abstract: The event data recorded by electromagnetic arresting gear exhibits high sampling rate, multi-dimensional coupling, and aperiodic transient, making it a complex non-stationary time series. In this paper, a similarity search method based on cascading lower bound pruning and global constraint early-abandoned strategy of DTW(dynamic time warping) is proposed to mine the event data for efficient fault diagnosis. Firstly, an improved piecewise aggregation approximation algorithm based on wavelet entropy and important point filtering is proposed to compress and reduce the dimension of event data. Secondly, three cascaded DTW lower bound distance functions are sequentially employed for pruning, eliminating dissimilar sequences and generating candidate sequences. Finally, a precise search is performed within the candidate sequence, optimizing DTW measurement speed using the proposed global constraint-based early-abandoned strategy. Experiments were carried out in the historical test data of the prototype of the electromagnetic arresting gear, which verified that the method proposed in this paper can match the working conditions and determine the system fault category accurately, and the efficiency is improved by 2.14 times compared with the mainstream TS2BC algorithm.