The invention discloses a typical scene generation method for a power distribution network power system, and the method mainly comprises the following steps: carrying out the normalization processingof power grid original data according to a time scale, and obtaining a clustering sample; preprocessing the clustering samples by adopting a density extreme value method to obtain an initial clustering center; according to the initial clustering center, adopting k-mean value clustering method to carry out synchronous clustering on wind power generation, photovoltaic power generation, tidal power generation and time sequence loads, and forming typical operation scenes of the power distribution network under different time scales. K-mean value clustering is improved by using density peak value clustering, problems that an initial clustering center and the number of clusters cannot be determined in traditional K-mean value clustering are solved, and compared with a traditional K-mean value clustering method, the improved algorithm is higher in clustering effectiveness and convergence.