Ensemble Automatic Sample Selector With Soft Actor–Critic for Imbalanced Fault Diagnosis

Aiming at the problem of sample selection of extremely imbalanced data for fault diagnosis in industrial processes, an ensemble automatic sample selector (EASS) based on the soft actor–critic (SAC) is proposed in this article. In EASS, SAC is employed to solve the decision problem of sample selection, and the idea of ensemble learning is introduced to improve the stability of the model. In addition, a new type of state (error density) is designed as its input, which plays the role of boosting in combination with ensemble learning and solves the problem of overfitting. The practicability and superiority of the proposed method have been verified on synthetic datasets and industrial datasets. Experimental results show that the proposed EASS method can effectively deal with the imbalanced modeling problem and improve the fault diagnosis accuracy and model generalization ability compared with other state-of-the-art methods.