The invention provides a photovoltaic power generation system maximum power tracking method and apparatus considering the factor of haze. The method includes the following steps: a. acquiring photovoltaic cell open-circuit voltage, short-circuit current, and cell temperature, and PM 2.5 concentration in air; b. inputting acquired data into a BP neural network prediction model improved based on a harmony search algorithm to obtain a MPP voltage predicted value; c. adjusting the PWM control signal duty factor of a DC-DC converter through a PI controller to enable the actual work voltage of the photovoltaic cell to reach the MPP voltage predicted value; and d. taking the MPP voltage predicted value as an initial value and tracking the maximum power of the photovoltaic cell at a specified disturbance step by means of a perturbation and observation method. The invention further provides a corresponding tracking device. Through the combination of the perturbation and observation method and the model forecasting method and the full consideration of the influence of haze weather to photovoltaic generation, the MPPT tracking precision can be effectively improved, the power loss of a static state is reduced, and the MPPT tracking speed can be improved.