An Oversampling Fault Detection Method for Unbalanced Industrial Process Data Based on Improved SVM

In order to solve the problem of unbalanced high-dimensional data in industrial process data, this paper proposes a sampling dimensionality reduction method. Firstly, the dataset is oversampled by a safe-level borderline SMOTE method, the number of fault samples is increased. Dimentionality of the data is reduced by the convolutional neural network with attention mechanism, and finally the fault detection of industrial process data is carried out through the parameters of the support vector machine optimized by Harris Eagle algorithm. The advantages of the algorithm are verified by comparing with several common machine learning algorithms in the relevant indicators of fault detection.