A heterogeneous ensemble learning diagnosis scheme for limited fault sample scenarios in transmission lines

In recent years, the transmission line fault diagnosis method based on machine learning has yielded remarkable outcomes. Nevertheless, in practical engineering, the fault sample size is typically small. This circumstance poses a challenge to the current intelligent diagnosis models in terms of difficult training. To address the issues of insufficient model training and low fault diagnosis accuracy resulting from the weak cognitive ability of fault data within a limited number of samples, a supervised learning model based on heterogeneous ensemble learning and the Weighted Average Soft Voting mechanism (HWM model) was put forward. By comprehensively assessing the strengths and weaknesses of each base model, such as the Gaussian process classifier (GPC), support vector machine (SVM), logistic regression (LR), and adaptive boosting classifier (AB), the framework assigns greater weights to models with higher confidence. The intention is to integrate the merits of each model and balance the overall complexity of the model, thereby attaining better classification performance. Experimental verification indicates that the proposed method can further enhance the fault diagnosis accuracy of the machine learning model fault diagnosis scheme under a limited number of samples. With a minimum sample size of 50 samples for all types of fault samples, the accuracy can still reach 93.84%, which is 8.13%-32.38% higher than that of the four included classifiers. Moreover, comparison with other deep learning models demonstrates its superiority. Consequently, the proposed diagnosis strategy holds significant value for the application of fault diagnosis in engineering.