For reducing the number of neural network training data, according to the output of the photovoltaic array under local shadow conditions, the concept is proposed about basic shadow occlusion type, so that the neural network only train small amount of data and the voltage of MPP can be accurately predicted. Firstly, the actual PV array data used to test the effect of BP neural network training on the tracking of the voltage of MPP only with the basic shadow shading type. Then, built the MPPT simulation system of photovoltaic power generation system and compared the perturbation method, fixed voltage method and BP neural network combined with the perturbation method to track the MPP. Finally, the analysis shows that the BP neural network combined with the perturbation method trained by the basic shadow occlusion type can effectively track the MPP, that is, the basic shadow occlusion type can reduce the acquisition of training data when the neural network tracks multi-peak MPPT.