Research On Optical Power Prediction In Optical Communication Protection System Based On Machine Learning

With the vigorous development of optical fiber network,optical fiber transmission has been widely used in all fields of society and become the main artery of information transmission in today's society.Among them,optical power is the most important index to measure the stability of a running optical fiber line.In order to ensure the stable operation of optical fiber as much as possible,optical communication protection system arises at the historic moment.In the operation of optical fiber protection,the optical protection system mainly plays the function of signal warning.When the optical power level drops sharply,it can predict in advance and make corresponding response to protect the stable operation of optical fiber communication.In this paper,the method model of machine learning is mainly studied,and the prediction accuracy of optical power signal is deeply studied and the original prediction model is improved.In this paper,the prediction task of optical power is taken as the starting point,and the ARIMA prediction model based on wavelet decomposition is established.Firstly,the original optical power data is decomposed by wavelet decomposition,and the trend and random fluctuation characteristics in the original data are stripped away.Then,the decomposed data are respectively trained and predicted in the ARIMA model,and the predicted value is obtained.The experimental results show that the ARIMA model based on wavelet decomposition has a greater improvement in the prediction accuracy than the single ARIMA model.In view of the data characteristics after wavelet decomposition,the LSTM model is introduced to establish the LSTM-ARIMA prediction model,so that LSTM and ARIMA can respectively predict the high frequency trend data and low frequency detail data after decomposition.Finally,the prediction structure is reconstructed to get the final prediction results.The experimental results show that the LSTM-ARIMA prediction model has a greater improvement in the prediction accuracy than the single ARIMA model and the ARIMA model based on wavelet decomposition.