For continuous improvement of productivity, accurate, stable, and reliable control of temperature and humidity is important in industrial production. Accurate prediction of air temperature and humidity can improve the predictability and stability of air conditioning control systems. In this paper, based on the cloud database of industry settings, an improved prediction model based on backpropagation (BP) neural networks was established to forecast indoor air temperature (IT) and relative humidity (IH) every 10 min and 6–72 h in advance. The experimental building was in Chongqing, a typical humid, hot-summer, and cold-winter area in China. The test data were used to determine the optimal parameters of the neural network model. The experimental results showed that the IT and IH predictions by our model have strong correlations with the actual data, with the coefficients of determination being 0.9897 and 0.9778, respectively. Compared with other literature, our model was more effective in temperature prediction. The presented method can be used for the prediction and control of the indoor temperature and relative humidity in industrial production.