Accurate forecasting of solar power generation plays an important role in stabilizing power dispatch. Therefore, numerous machine learning models and deep learning models have been used for power generation forecasting. However, since a single independent model still has performance limitations, the prediction model of the ensemble model is adopted to aggregate the advantages of different independent models. Furthermore, its prediction error and generalization performance are both better than those of a single model. In this study, the stacking ensemble method was employed to gather four different base learners and then incorporated with the k-fold cross-validation for model training and testing. Meanwhile, a mobile IoT data collection system was also built, in which IoT sensors were applied to collect data of weather factors (solar radiation, ambient temperature, humidity, and wind velocity) as well as of solar power generation. Next, the real-time monitoring system for solar power generation developed in this study displayed real-time power generation and data storage. The experimental results showed that the root mean square error of the solar power generation prediction model, that is, the regression ensemble model (RGEM) proposed by this study, dropped by 6.24, 8.31, 9.94, and 4.21%, respectively, in model testing compared with those of the independent model support vector regression, least squares support vector regression, least absolute shrinkage and selection operator, and ridge. Besides, RGEM’s testing mean absolute percentage error = 0.0966 indicated a prediction model of high accuracy.