The invention discloses a photovoltaic power generation system maximum power tracking method based on model identification. The method comprises the steps that firstly, parameter identification is conducted on a photovoltaic array reference mathematic model by adopting measurement information of a practical photovoltaic power generation system, and an identification model consistent with the practical object is obtained; secondly, a BP neural network is constructed, a series of output voltage and output current test data of the model under different illumination intensities and photovoltaic cell panel temperatures is generated by adopting the identification model, information such as output voltage and output power corresponding to the illumination intensities, the photovoltaic cell panel temperatures and the maximum power points is extracted to be used for training the BP neural network, and parameters of the BP neural network are obtained; finally, the trained BP neural network is used for maximum power tracking of the photovoltaic power generation system on line, and therefore the real-time property and the high efficiency of photovoltaic power generation system maximum power tracking are greatly improved. According to the method, the principle is easily implemented on line, and the flexibility is good.