The invention provides a distributed photovoltaic ultra-short-term generated power prediction method and system, and belongs to the technical field of power dispatching. The prediction method comprises the following steps: S1, training based on sample data to obtain a distributed photovoltaic power prediction model; s2, according to real-time generated power and weather forecast data, utilizing the distributed photovoltaic power prediction model to carry out generated power prediction; the step S1 comprises the following steps: S11, acquiring sample data, wherein the sample data comprises historical power generation power and historical meteorological data; s12, performing normalization processing on the sample data; s13, establishing an LSSVM (least square support vector machine) prediction model; s14, optimizing kernel parameters of the LSSVM model by using a particle swarm algorithm; and S15, constructing a distributed photovoltaic power generation power prediction model. According to the invention, by constructing the generation power prediction model and the prediction system, distributed photovoltaic online prediction modeling and model online training optimization are realized, high-precision power prediction data are output, and the new energy grid-connected access and consumption management level is improved through auxiliary scheduling.