Abstract: Flow feature analysis is an important research area of fluid mechanics. Traditional feature analysis methods, which mainly based on magnitude of the flow parameter, is greatly affected by the parameter form and subjective threshold. In this paper, a new flow field feature analysis method is proposed based on flow time history deep learning, and flow time history feature extraction analysis framework based on autoencoder is established; The unsupervised training method is used to recognize the hidden complex features in the flow time history signal, and realize the low-dimensional representation and feature analysis of the complex time history features in the flow field. The flow field feature analysis of cylinder with Re_D = 200 was carried out, and the low-dimensional characterization of periodic laminar flow field data was obtained. Flow feature distribution was directly obtained according to the latent code, and results obtained proved to be reasonable. The proposed method can provide new methods and references for solving problems such as flow field feature extraction, flow feature analysis and flow feature representation.