High-resolution time variant flow field data is the key to the study of turbulence flow. Limited by measurement methods, simulation efficiency and data storage, it is still difficult to obtain high-resolution turbulent flow data directly in some circumstances. In this paper, based on the low-dimensional representation model of flow time-history data, a neural network-based feature coding prediction model and high-resolution turbulence flow reconstruction method were proposed. Firstly, a low-dimensional representation model of the turbulence flow was established based on the one-dimensional convolution networks; then, an artificial neural network model was employed to establish the mapping between the measuring point coordinates and feature coding system, and the prediction of feature coding for the unknown measuring points was realized; finally, based on feature coding, the decoder in the representation model was utilized to generate turbulence flow time history data at unknown positions. Turbulence flow with Re=2.2×10⁴ around a square cylinder was studied, and the low dimensional representation model and flow generation model were trained and verified. The method proposed in this paper is a high-precision turbulence flow data reconstruction method which can be widely used in one-point-based sensor data processing. It is a new approach for the reconstruction of turbulence flow field time-history data.