Abstract: With the wide application of hydro-photovoltaic（PV） complementary system, accurately describing the uncertainty of the hydro-PV output is of significance for the operation and planning of the power grid. The existing multi-source fusion characteristic modeling methods not only have the problem of insufficient consideration of the spatio-temporal correlation of renewable energy output, but also need to make prior assumptions in the practical complex application environment, which leads to the reduction of the generation quality. Based on this, a stochastic scenario generation method based on denoising diffusion probabilistic model（DDPM） for hydro-PV complementary system is proposed. Firstly, the difference between the generated noise and the original noise distribution is measured by combining Euclianian distance and L2 regularization loss function, and the UNet network structure suitable for the hydro-PV-load stochastic scenario generation is designed. Then, by continuously adding noise to the forward process and denoising training during the backward process, the changes of the correlation and the fluctuation characteristics of hydro-PV-load multi-dimensional variables are captured, and their probability distribution is fitted. Finally, the collaborative modeling of multi-source data is carried out to efficiently generate the hyrdo-PV-load stochastic scenario. A case is used to test the actual data collected from a regional power grid, and the effectiveness of the proposed method is verified by comprehensive evaluation indices.