Fault Classification Method based on Self-paced Neighborhood Preserving Embedding

In view of the large amount of faulty data in industrial systems, strong correlation between data, large redundancy, and high dimensionality, traditional classification methods can easily lead to the phenomenon of "dimensionality disaster". On the basis of the Neighborhood Preserving Embedding algorithm (NPE), a fault classification algorithm based on Self-paced NPE is proposed. First, the neighbor point set and adjacency matrix of the fault sample points are obtained through the NPE algorithm. Secondly, the function of solving the projection matrix in the NPE algorithm is used as the loss function in the Self-paced Learning model to construct the objective function of the Self-paced Neighborhood Preserving Embedding (Self-paced NPE) algorithm. By designing the loss function in the Self-paced Learning model, gradually select from easy to difficult samples with smaller loss are selected, samples with larger loss are discarded, and relatively simple samples are used for training. Finally, the objective function of the Self-paced NPE algorithm is solved and the proposed algorithm is tested with TE process data . The experimental results show that, compared with the NPE algorithm, the Self-paced NPE algorithm can extract features of high-dimensional fault data more effectively, achieve dimension reduction, and improve the accuracy of fault classification.