Abstract:To overcome the temperature error problem of F-P filter in FBG sensing system, a temperature error compensation method for F-P filter based on improved DNN is proposed. Adding BN layer, Dropout layer, attention mechanism layer, and residual connection layer on the basis of traditional DNN model improves the performance of DNN model; By using the improved DNN model to predict the wavelength drift of F-P filters at different temperatures and voltages, and then converting the wavelength of F-P filters at different temperatures to the wavelength at a fixed temperature, temperature errors are eliminated. An experimental system is built to experimentally verify the proposed method, and compared with the prediction results of LSTM and LSTM+CNN error compensation models. The experimental results show that the maximum standard deviation of the wavelength after error compensation is 4.7 pm, and the maximum fluctuation from the average is 13.3 pm. Compared with before compensation, the standard deviation is reduced by 99.96%, and the maximum fluctuation phase is reduced by 99.93%. Using the method proposed in this paper, the accuracy of FBG sensing can be significantly improved without adding additional hardware, and the complexity and cost of the system can be reduced.