In a class of photoelectric tracking systems with only the target image sensor and MEMS accelerometer installed, an equivalent acceleration feedforward control method can improve the tracking ability of the system effectively. However, due to the low-frequency noise of accelerometers, the delay of the synthetic trajectory and the uncertainty of the movement model, the tracking accuracy will be limited. Therefore, in this paper, an equivalent acceleration feedforward method based on sensor optimization and robust prediction is proposed to further improve the tracking ability of the system. The frequency-domain fusion using the accelerometer measurement and the calculated acceleration can optimize the low-frequency performance of the accelerometer. Meanwhile, the robust prediction algorithm can reduce the impact of the delay of the synthetic trajectory and the uncertainty of the movement model and give more accurate feedforward acceleration. The experiment shows that this method can further improve the tracking ability of 0.1 Hz~ 4.5 Hz.