The optical signal to noise ratio(OSNR) is one of the key indexes for health assessment and performance prediction of optical network. However, traditional methods face great challenges for computing accusation and system robustness. Based on the recurrent neural network and attention mechanism, this paper presents an OSNR calculation method based on attention mechanism. This method directly captures the logical relationship between the timing input and the OSNR, and improves the calculation accuracy of OSNR. According to different fiber input power, transmission distance and demodulation methods, a variety of experiments are designed. The results show that the proposed method can reduce the calculation error of OSNR to less than 0.360 dB.