The invention discloses a data analysis method for photovoltaic power generation output prediction. The method specifically comprises steps as follows: historical data of power generation output in a certain area and corresponding influence factor data are acquired; a historical output curve is decomposed into a historical subject output curve A3 and three historical detail output curves D1, D2 and D3 through wavelet transformation; four forecast sub-models are established respectively for the historical subject output curve A3 and the three historical detail output curves D1, D2 and D3 by use of a PSO-RBF (particle swarm optimization-radial basis function) neural network; a subject output curve shown in the specification and three detail output curves shown in the specifications of a to-be-predicted day are calculated respectively, and a prediction value of the output curve of the to-be-predicted day is obtained through wavelet reconstruction of the subject output curve and the three detail output curves of the to-be-predicted day. According to the method, wavelet transformation and the PSO-RBF neural network are combined together, and the method has higher use value for formulation of power generation schedules through cooperation of a photovoltaic power generation system with a power system.