摘要：For predicting the conversion rate of vinylchloride monomer (VCM) in the polyvinyl chloride (PVC)production process, a neural network soft-sensor model basedon data dimensionality reduction strategies was proposed. Inorder to solve the problem of complex neural network topologyand long training time caused by the excessive input vectordimension, seven kinds of data dimensionality reductionmethods, such as principal component analysis (PCA), localitypreserving projection (LPP), kernel principal componentanalysis (KPCA), expectation max principal componentanalysis (EMPCA), local tangent space alignment (LTSA),T-distributed stochastic neighbor embedding (TSNE) andneighboring preserving embedding (NPE), are used to reducethe dimension of the high-dimensional input data used in theneural network soft-sensor model. Then the radial basisfunction (RBF) neural network based on the gradient learning,orthogonal least squares and clustering learning methods andthe dynamic fuzzy neural network (D-FNN) were utilized torealize the prediction on the VCM conversion rate. Simulationresults show that the proposed neural network soft-sensormodels based on seven data dimensionality reduction strategiescan effectively predict the key economic and technicalindicators of PVC polymerization process and meet thereal-time control requirements of PVC production process.
