Enhanced Fault Localization in Multi-Terminal HVDC Systems Using Improved Gaussian Process Regression

Abstract

Accurate fault localization is crucial for protecting DC networks following the successful detection of internal faults in power transmission systems, as it minimizes the need for replacements and enables swift power recovery. This study proposes an improved Gaussian process regression (IGPR)-based fault location method for a multi-terminal high voltage DC (HVDC) system. Initially, DC voltage and current signals are captured from relay locations (both positive and negative poles) under various operating conditions during a DC fault event. Subsequently, twelve statistical features (such as mean, median, and standard deviation of voltage and current signals) along with two additional features based on the coefficient of correlations between voltage and current are extracted. These features serve as inputs for training the proposed IGPR model, followed by random sample testing. The IGPR model incorporates hyperparameter tuning of Gaussian process regression using Bayesian optimization. Additionally, a detailed analysis on feature selection is conducted using the RRelief feature selection method. Performance evaluation of the proposed method is carried out by comparing it with several other improved machine learning models and existing literature. The results demonstrate that the proposed approach outperforms others, achieving a very low root-mean-square error of 0.004 with a coefficient of determination (R2) of 0.9956.

Keywords

• Artificial intelligence
• optimization algorithm
• Gaussian process regression
• HVDC
• multi-terminal
• hyperparameter tuning