The invention discloses a photovoltaic power prediction method in combination with a photovoltaic power physical model and data driving, belonging to the field of new energy prediction technology of power systems. The method comprises the following steps: determining key weather features that affect the photovoltaic power by using a photovoltaic power physical model, and establishing key weather feature matrices of a historical period and a prediction period; and then separately establishing weather data matrices of the historical period and the prediction period to obtain input matrices of the historical period and the prediction period; performing feature extraction for the input matrices to obtain principal component feature matrices of the historical period and the prediction period; and selecting K historical periods with the nearest Manhattan distance from the principal component features of any prediction period, fitting to obtain a mapping relationship between the principal component features of the K historical periods and the photovoltaic power of the corresponding historical periods, and inputting the principal component features of the selected prediction period into the mapping relationship to obtain the photovoltaic power of the prediction period. According to the photovoltaic power prediction method disclosed by the invention, the photovoltaic power can be accurately predicted by using the photovoltaic power physical model, and stronger industrial application values can be achieved.