The invention discloses a period division-considered nonlinear photovoltaic power generation output forecasting method. The method comprises two steps of: establishing a division model and a nonlinear time sequence model to forecast solar radiation intensity, and establishing a photovoltaic generated power model: firstly carrying out deduction via a maximum posterior probability to obtain the division model, wherein the division model takes practical engineering experiences as radiation intensity prior information and is used for dividing nonstationary time sequences by considering distribution characteristics of radiation intensity sequences, and each sequence after the division is stationary sequence so that the assumption that nonlinear SIAVAR single-index addible vector auto-regression models require the sequences to be stationary is solved; simulating each sequence by using a nonlinear SIAVAR model; carrying out global approximation on non-parameter functions in the nonlinear model through nonlinear simulation, and carrying out semi-parameter estimation by using a Backfitting method so as to complete radiation intensity time sequence fitting; and establishing a photovoltaic power generation output model according to a relationship between photovoltaic power generation output and the solar radiation intensity, so as to obtain a photovoltaic power generation output forecasting time sequence.