To meet the wide access of distributed photovoltaic （PV） in distribution network, this paper presents a generating method for spatiotemporal probability distribution of distributed PV output based on Latin hypercube sampling. In this paper, after the factors that affect PV output are analysed, FFT filtering combined with second-order difference screening is used to decompose PV output into basic values and fluctuation values considering the similarity of the external environment in a small range, which improves the traditional method. The correlation is considered to exist only between the fluctuation values in a small range instead of the whole output values of the traditional method. Further the relationship between the correlation coefficient and the distance is further established by the correlation coefficient matrix fitting method. The Lilliefors test process is added before the sampling space transformation in the Nataf transform to determine whether the distribution of the fluctuation value satisfies the normal distribution. Complex calculation caused by the spatial transformation can be avoided and the multivariate joint distribution can be constructed quickly if the distribution of fluctuation value satisfies the normal distribution. Finally, by using the PV output simulation generating model, the simulated probability distribution of PV output can satisfy the probability distribution of single PV output in time series and fit the probability distribution PVs output simulation of the correlation coefficient matrix in space.