Accurate prediction of heliostat temperature was the key to achieve efficient greenhouse regulation, which was of great importance to crop growth and development, but it was difficult to achieve continuous and accurate prediction due to the characteristics of time series, nonlinearity and multi coupling of temperature. At the same time, the current production regulation of greenhouse mostly depended on the relevant experience of producers. This method had caused the lag of feedback control and affected the growth of crops.A temperature prediction model of solar greenhouse based on 1D CNN-GRU was proposed. The internal and external environmental factors were obtained through the monitoring platform inside and outside the greenhouse, and the strong correlation features and structural features were obtained by Spearman correlation coefficient and the two-dimensional matrix input network with time step, which was used for temperature prediction. The determination coefficient of the model after 1~4h prediction on the test set was 0.970~0.994, the root mean square error was 0.612~1.358℃, the average error was 0.428~0.854℃,and the maximum absolute error after the absolute value was 0.856~1.959℃. The model was verified under different KT and the results showed that the model had the best prediction effect when KT≥0.5(sunny), and the model also achieved ideal prediction accuracy under other KT, indicating that the model was universal and provided an important basis for accurate and efficient temperature control of solar greenhouse.