Solar energy has the advantages of being clean, safe, and renewable, and photovoltaic (PV) power generation can reduce resource consumption and contribute to sustainable development. However, PV power is easily affected by weather, and the prediction of PV power for different weather types is also a research difficulty. This study proceeds to apply synthetic minority over-sampling technique (SMOTE) and machine learning for PV power prediction under different weather types. Firstly, the meteorological factors that have the greatest impact on PV power are selected by the Pearson's correlation coefficient method. Then the sunshine duration is calculated based on the meteorological factors with a greater degree of importance, and the weather is classified as sunny, cloudy or cloudy, and snow-covered days by setting a threshold for the number of hours of sunshine, and then the samples under various weather types are expanded by the SMOTE technique. Finally, the PV power prediction model is constructed by various machine learning algorithms for different weather scenarios and before and after data expansion. Through case validation, it can be seen that the algorithm proposed in this paper is able to classify different weather types, and provides a solution to the sample imbalance problem of PV power prediction under different weather types, which improves the prediction accuracy of PV power under different weather scenarios.