The invention relates to a photovoltaic ultra-short-term power prediction method based on an LM-BP neural network. The method includes the following steps that an LM-BP neural network prediction model is built according to environmental factors, and input layer data and output layer data are determined; the LM-BP neural network prediction model carries out normalization processing on input data; LM-BP neural network trainings are carried out by means of an LM algorithm, and an LM algorithm with an updated weight is obtained; photovoltaic power of a photovoltaic power station is calculated to obtain a prediction result. According to the photovoltaic ultra-short-term power prediction method, the LM algorithm and a BP neural network are combined, the LM-BP neural network prediction model is built on the basis of a BP neural network prediction structure, photovoltaic ultra-short-term power prediction is carried out by means of the LM algorithm so that any continuous nonlinear function can be approximated at any precision, the dynamic random problem of photovoltaic power prediction can be quickly and accurately solved, prediction speed and prediction precision are improved, and the method has good robustness.