Although the data-independent acquisition (DIA) technology and the related algorithms are developing rapidly, the results of DIA data analyzing without the use of spectra library from data-dependent acquisition (DDA) data remain unsatisfying. Here we proposed an untargeted analysis method, Dear-DIA-XMBD, for directly analyzing DIA data. Dear-DIA-XMBD first integrates the deep variational autoencoder and triplet loss to learn the representations of extracted ion chromatograms of fragments, then uses the k-means clustering algorithm to aggregate fragments with similar representations into the same classes, and finally establishes the inverted index tables to determine the precursor of fragment cluster between precursors and peptides, and between fragments and peptides. We show that Dear-DIA-XMBD performs superiorly with the highly complicated DIA data of different species obtained by different instrument platforms. We trained the Dear-DIA-XMBD on an Escherichia coli SWATH dataset. We benchmarked the performance of Dear-DIA-XMBD by using the highly complicated sample datasets, which consist of L929 mouse dataset, SWATH-MS Gold Standard (SGS) human dataset (PeptideAtlas PASS00289), HYE124 dataset (PXD002952) with 64 variable windows (TripleTOF 6600).