Abstract: Nutrient Film Technique (NFT) has been one of the most important production modes of hydroponically grown lettuce in the north of China. However, the low temperature in winter and high temperature in summer can be two of the key issues to restrict the sustainable development of hydroponically grown lettuce. In this study, an acquisition monitor platform of temperature data was constructed inside and outside the greenhouse using sensors combined with the Internet of Things (IoT), in order to reduce the delayed temperature control of lettuce in the root zone in winter and summer. A specific measurement was performed on the indoor air temperature, indoor soil temperature, indoor air humidity, nutrient solution temperature, nutrient solution temperature at the return port, outdoor carbon dioxide concentration, outdoor solar radiation illumination, outdoor air pressure, outdoor air humidity, outdoor air temperature, and outdoor wind speed. The air temperature and outdoor wind speed were also collected. Pearson correlation analysis was used to determine the correlation between the ambient environmental factors and the root zone temperature after 30 min. Outdoor air pressure and outdoor air humidity shared a weak correlation with the root zone temperature after 30 min. The rest of the influencing factors were highly correlated with the root zone temperature and thus were selected as influencing features. The prediction model of root zone temperature was constructed using BP neural network and machine learning, according to the historical environmental data inside and outside the greenhouse. The optimal number of hidden layers was experimentally determined for the BP neural network, which was 15 in summer and 13 in winter. The input weights and thresholds of the BP neural network model were optimized to improve the accuracy of the BP model using dung beetle optimizer (DBO). The optimal weights and thresholds were obtained through the positional change of the dung beetles using five operations, namely, ball rolling, foraging, stealing, and reproducing. Ultimately, the DBO-BP neural network was constructed in the winter and summer seasons. The prediction model of root zone temperature in the cultivation tank was constructed to compare with GA-BP and BP neural network models. The results showed that there were more consistent trends of the predicted and real root zone temperatures. The maximum error of the temperature prediction of the DBO-BP model in winter was 2.36 °C with a coefficient of determination of 0.933, and the maximum error of the temperature prediction of this model in summer was 2.21 °C with a coefficient of determination of 0.943, while the coefficients of determination of the GA-BP and the BP models in summer were 0.928 and 0.892, respectively. The root mean square error and the average absolute error of the DBO-BP model evaluation indexes were 0.707 and 0.549, respectively, which were smaller than the rest. Thus, the DBO-BP neural network can fully meet the demand for the temperature prediction accuracy of the root zone in NFT cultivation. The finding can also provide an effective mode for rapid temperature control in the root zone of lettuce cultivation.