The invention discloses a short-term photovoltaic power prediction method based on an improved EMD algorithm and an Elman algorithm. The short-term photovoltaic power prediction method comprises the steps: step S1, carrying out clustering analysis on historical data, and determining a category of a to-be-predicted day and corresponding irradiation intensity to-be-predicted time intervals; step S2, establishing a same-type day time sequence in the category of the to-be-predicted day according to main environmental characteristics; step S3, utilizing the improved EMD algorithm to perform median filtering on the same-type day time sequence, carrying out mode decomposition according to fluctuation degrees, and classifying same-type modes into a category; step S4, adopting the Elman algorithm to predict irradiation intensity of each mode category, and further acquiring photovoltaic hourly power generating power value. The short-term photovoltaic power prediction method aims to increase prediction precision of irradiation intensity under the condition of weak irradiation, is proven to be adapt to irradiation intensity prediction of different-type days, and achieves more rapid and accurate prediction.