KELM transmission line fault location method incorporating uncertainty estimation

Abstract

Abstract With the widespread application of flexible direct current transmission (MMC-HVDC) in the interconnection of large power grids and the consumption of new energy, the accuracy, reliability, and response speed of transmission line fault location are of crucial importance for the safe operation of the system. To address the issues of noise sensitivity and difficulty in characterizing uncertainty in traditional traveling wave localization methods, the paper proposes a kernel extreme learning machine (KELM) fault localization method that introduces uncertainty estimation. This method first constructs a ± 500 kV MMC-HVDC system simulation model on the PSCAD/EMTDC platform, collects traveling wave signals under different fault conditions, and extracts time-frequency features using wavelet denoising and fixed parameter VMD techniques; Subsequently, a KELM model based on BWO (Beluga Whale Optimization) optimization was constructed to globally optimize the kernel width σ and penalty coefficient C, achieving high-precision regression mapping. In order to further enhance the engineering interpretability of the results, an uncertainty estimation module is introduced to provide confidence intervals and credibility evaluations for the predicted output. The simulation results show that the proposed method has an average absolute error (MAE) of only 0.19 km and a 95% confidence interval coverage rate of 95.8%. Under complex operating conditions such as low signal-to-noise ratio (30 dB) and high transition resistance (300Ω), the positioning error is still controlled within 0.31 km, which is significantly better than traditional ELM, KELM, and PSO-KELM methods, verifying the accuracy, robustness, and practicality of the proposed method.