Implementation of Long Short Term Memory Method to Predict Fiber Optic Network Disruptions

As it known, fiber optic cables are widely used by internet and telecommunications service providers to transmit images, voice messages and data. However, it cannot be denied that using fiber optic cables has also experienced interference with the Fiber To The Home (FTTH) architecture. Network disturbance factors are caused by external factors such as lightning strikes and fallen trees on Optical Distribution Point (ODP) devices on poles, construction projects, and even pests, causing equipment damage, cable bending and cable breaks. Damaged devices and broken cables cause loss or loss of internet connection. This disturbance causes loss which is indicated by the total attenuation value outside the standard limit (-13 to (-24) dB). This problem causes users not being able to access the internet and it took a long time for maintenance and repairs in the field. This study uses the Long Short Term Memory (LSTM) method which is part of deep learning to predict the future. The implementation process was carried out by entering datasets, dividing training data and testing data, building system models, training and testing processes, and evaluating prediction results using Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE). From the conducted research, it is predicted that in 312 data there will be 14 disturbances from 13 - 28 June 2023. Comparison between the LSTM method and the Fuzzy Time Series Model algorithm obtained MAPE and RMSE values of 0.007% and 0.37 for LSTM. However, Fuzzy Time Series were 0.64% and 0.11. So the percentage of success of the LSTM method for MAPE and RMSE is 99.99% and 99.63%. Hence, the LSTM method is feasible in predicting time series data compared to the Fuzzy Time Series Algorithm.