Fault ranging of 750kV UHV AC line based on joint self-supervised learning

During travelling wave propagation, complex refraction and reflection of reflected signals from complex distribution lines occurs, resulting in the principle of single-ended travelling wave fault ranging being unsuitable for 750kV UHV AC scenarios, the forward wave being difficult to effectively correspond to fault sections with different fault distances, the existence of poorer sample fitting ability, and the lower accuracy of actual ranging results, and the proposal of a joint self-supervised learning based fault ranging method. By constructing a state space model of the fault signal of UHV AC line, the dynamic relationship between each electrical component and state variables in the line fault is described; based on the analysis of the traveling wave reflection of the fault voltage of UHV AC line, the time of the forward wave arriving at the fault node is calculated, and the fault section of the UHV AC line is identified; based on the joint self-supervised learning method, the denoised self-coder is improved by using the deep learning framework to analyse the fault section at different fault distances. Based on the joint self-supervised learning method, a deep learning framework is used to improve the denoising self-encoder to analyse the degree of each electrical component and state variable corresponding to different fault segments at different fault distances, and to achieve the fault ranging of UHV AC lines of 750kV. The experimental results show that under different fault distances, the measurement error between the fault ranging results and the actual distance of UHV AC line is 1.15km at the highest value, and the minimum value is only 0.18km, and the lowest value of the average running practice in the training is 7.02s. The method has a strong fitting ability for the set of small samples, and it can effectively maintain a high computational efficiency, and at the same time, it can effectively locate the fault segments, and it can be applied to the fault segments generated by different fault distances. Different fault distances produce more accurate actual ranging results.