A method for judging the validity of audible noise measurement data of high-voltage DC lines under finite labeled samples

When measuring the audible noise of high-voltage direct current lines outdoors, many invalid data are often measured due to the interference of external environmental noise, causing difficulties for later data analysis. This paper presents a method for determining the validity of audible noise measurement data of high-voltage direct current lines under a finite label sample. First, the Transformer model is used to extract the time-frequency characteristics of the audible noise measured signal; Then, each feature vector is treated as a node to build a graph network, and label propagation is carried out based on the similarity between nodes when the validity of a small number of samples is known, thereby determining the data validity of the remaining samples; Finally, the audible noise measurement data of a certain high-voltage direct current line was analyzed, and the results showed that the method had an average accuracy of 99.0%, an average recall rate of 94.0%, and an average f1 score of 0.966, all of which were superior to several common machine learning models such as logistic regression, SVM, and random forest. In the case of known validity of a small amount of data, the algorithm significantly improves the accuracy of invalid data detection, which can provide great assistance in judging the validity of audible noise measurement data.