AI based failure analysis for one type of cables in power systems

Abstract

The dissertation showcases research conducted on the utilization of artificial intelligence (AI) techniques for analyzing power cable faults. To develop the AI models, a dataset of power cable fault patterns was created using the Matlab/Simulink platform, which allowed dynamic simulation of the fault model. Next, the dataset was partitioned into training, validation, and testing sets, underwent pre-processing, and was employed to train six distinct AI models, comprising three deep learning models (LSTM, 1D-CNN, and 2D-CNN) and three machine learning models (SVM, KNN, and Random Forest). Subsequently, the trained models were evaluated on the test set to gauge their effectiveness in dealing with unfamiliar data. The results showed that all models achieved high accuracy and recall rates in predicting the defect patterns, with the 2D-CNN model performing best among the deep learning models and the KNN classifier performing best among the machine learning models. However, all models made errors in predicting open circuit faults, which are difficult to distinguish from other faults due to their similar characteristics in different phases.

The study discusses future directions in AI for power cable fault analysis, including improving data quality, cloud/edge computing, interpretability/traceability, and integrating AI with engineering practices. The study highlights the potential of AI to improve the safety/stability of power systems through reliable fault analysis.