基于DNPE-SVDD的化工过程监控

Chemical processes test dataset are high dimensional, and have the combined feature of nonlinear and dynamic characteristics. However, traditional linear dimension reduction algorithm cannot extract the local structure information and dynamic characteristics. The monitoring model of chemical process based on dynamic neighborhood preserving embedding-support vector data description (DNPE-SVDD) algorithm is proposed. With the superiority of NPE in nonlinear dimensionality reduction and SVDD in the detection of outliers, dimension is reduced by DNPE algorithm and the monitoring model of the manifold space with reduced dimension is established by SVDD algorithm. The Tennessee Eastman (TE) process is simulated using the proposed model. Compared with DPCA and DNPE algorithms, the simulation results show that the DNPE-SVDD has a higher accuracy of fault detection.