Optical signal-to-noise ratio (OSNR) is closely related to the transmission performance of optical fiber communication, so OSNR monitoring is a crucial part of optical performance monitoring technology. At the same time, the dispersion in the transmission link will lead to the broadening of optical signal pulses, which will reduce the accuracy of OSNR monitoring. Aiming at this problem, a convolutional neural network model is designed. The asynchronous delay sampling graph (ADTP) is used as the network input feature, and the instance batch standardization module is introduced. It inherits the advantages of feature divergence distribution at different depths of the neural network and improves the adaptability of the neural network to dispersion changes. The experimental results show that the mean absolute error (MAE) of the model is 0.2 dB in the case of 10 Gb/s NRZ-OOK signal without dispersion interference monitoring, and the MAE is reduced by 0.61 dB at most in the case of dispersion change.