High Impedance Fault Detection in Resonant-Grounded Distribution Networks Using Continuous Wavelet Transform and Deep Residual Networks

In the resonant grounded distribution network, High Impedance Fault (HIF) exhibits subtle and fragile features that are often obscured by normal disturbances, making them difficult to detect accurately. Thus, this paper presents an innovative HIF detection method that combines Continuous Wavelet Transform (CWT) with Deep Residual Networks (DRN) to enhance detection accuracy and adaptability based on measuring zero-sequence current. By thoroughly analyzing the zero-sequence current characteristics, we discover that the time-frequency domain distribution of these currents more effectively captures the deep features of HIFs. Based on this insight, we propose a CWT-based feature extraction technique to highlight these distinguishing features. Building upon this, we develop an improved DRN-based HIF identification model that not only explores the deep properties of HIFs but also addresses issues of gradient vanishing and performance degradation commonly found in traditional CNNs. Extensive numerical simulation testing and comparison conducted using PSCAD demonstrate that the proposed HIF detection method significantly outperforms existing techniques in terms of accuracy and adaptability. And the results of hardware-in-the-loop experimental testing validate the effectiveness of our approach in reliably detecting HIFs, thereby contributing a valuable advancement to the field of power system protection.