Soil temperature is one of the important factors affecting crop growth and greenhouse thermal environment. In order to accurately predict the greenhouse soil temperature, a one-dimensional unsteady heat transfer model of soil was constructed by using theory of computational fluid dynamics. In order to solve the difficult problem of obtaining boundary condition, a boundary condition prediction model was constructed by using long short term memory (LSTM) neural network. Accuracy verification test of one-dimensional unsteady soil heat transfer model was firstly carried out with boundary condition measured by sensor. The results showed that the variation trend of calculated and measured soil temperature in different seasons and depths were consistent. The maximum value of mean absolute error (MAE) and max absolute error (MaxAE) between predicted and measured soil temperature were 1.29℃ and 2.16℃, respectively. Secondly, the prediction model of boundary condition was verified. The results showed that the determination coefficient (R2) between predicted and measured value of boundary condition was 0.99. The maximum value of MAE and MaxAE between predicted and measured value of boundary condition were 0.18℃ and 2.63℃, respectively. The results indicated that the model could accurately predict the boundary condition. Finally, the predicted and measured boundary condition data were introduced into one-dimensional unsteady heat transfer model of soil, respectively. The calculated results of the model with measured and predicted boundary condition were compared with the measured soil data. The results showed that the simulation results of soil temperature with predicted and measured boundary condition were consistent. The maximum deviation of R2, MAE and MaxAE between calculated and measured soil temperature under measured and calculated soil boundary condition were 0.03, 0.14℃ and 0.92℃, respectively. The above results showed that one-dimensional unsteady heat transfer model of soil under predicted boundary condition can predict the soil temperature in different depths accurately.