Advanced Artificial Intelligence Techniques for Fault Distance Prediction in Optical Fibres

It is necessary to estimate the value of distances to faults in the optical fibres for proper functioning and fault diagnosis of the fibre optic systems. This research proposes a comparison of result outcomes within numerous categories of machine learning algorithms such as Decision Tree, Random Forest, Gradient Boosting, XGBoost, LightGBM, Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM) on a new and emerging Fibre Optic Fault Distance Dataset. Given dataset contains sequenced OTDR signatures corresponding to different types as well as positions of the faults. The data then went through data pre-processing and was separated into training and test sets where models were trained from 80% of the data and tested on 20%. Model accuracy was also determined by the commonly used performance parameters including R-Square and Mean Squared Error (MSE). They found out that the Random Forest and LSTM models represented the highest R-Square values and the lowest MSE, thus proving the algorithm’s capability to forecast the fault distances. The greatest results were obtained by the proposed hybrid LSTM–RF model which combines the sequence processing and ensemble learning. These studies show that effective diagnosis of optical fibre faults using higher level and more complex techniques in machine learning is possible and specify directions on further study and application of this subject.