When the BP neural network is adopted to predict the voltage at the maximum power point, there is a big error if the light intensity changes drastically. Aiming at this problem, a novel improved fruit fly optimization algorithm(IFOA) determining the optimal BP neural network parameters (weight and threshold) is proposed, and a simulation model of the photovoltaic system MPPT control strategy based on the IFOA-BP neural network algorithm was established. The test and simulation results show that, IFOA has a great advantage in search speed and accuracy than FOA; IFOA-BP neural network can effectively increases the convergence speed and reduces the prediction error; compared with the incremental conductance(INC) method, the proposed photovoltaic system MPPT control algorithm based on IFOA-BP neural network could suppress the oscillation around the maximum power point(MPP) under steady-state conditions and track down the MPP quickly and accurately when light intensity and temperature change drastically, which verifies the stability, precision and rapidity of the proposed MPPT method.