Study On Data Based Soft Sensing Methods And Applications

Soft-sensing technique is able to estimate the difficult-to-measure primary variables from the easy-to-measure secondary variables by building some mathematical models. It has been one of the most attractive research topics in the area of process control in this decade. There are two kinds of soft sensing techniques: mechanism based and data based. Data based soft sensing technique encounters the following problems: large number of highly correlated variables, limited data samples, highly nonlinear and time-varying process, etc. This dissertation focuses on the research of partial least squares (PLS) method and addresses the above problems successfully. Main results and contributions of this dissertation are as follows: 1. New geographical explanation is given for both PLS and principal component regression (PCR) methods. It has been proved that PLS model can give better fitness performance than the PCR model with the same number of latent vectors. An improved orthogonal signal correction (OSC) method is proposed as a preprocessing procedure of PLS method to remove from X (independent variables) the variation information which is not correlated to Y (dependent variables). The resulted model has reduced complexity and improved interpretation ability. 2. A novel nonlinear PLS (NLPLS) model, named RBFPLS, is proposed by combining PLS with RBF networks to deal with the high nonlinearity of the process. Simulation results show that the RBFPLS model has good prediction performance. To adapt process changes, a recursive nonlinear PLS (RNPLS) algorithm is proposed to update the RBFPLS model and the complete on-line model updating procedure is given. The experiment results demonstrate the effectiveness of this algorithm. 3. RBFPLS method and RNPLS algorithm are applied to the optimization of batch processes and a batch-to-batch optimization approach based on recursively updated RBFPLS models is proposed. By applying this optimization method, the final product quality can converge to satisfactory result smoothly and quickly within only a few batches. The simulation result shows that the proposed method achieves better performance than that with a PLS model. 4. An operating regime based soft sensing model of polypropylene melt flow rate (MFR) is developed and implemented on the actual plant to provide operation guidance. The accuracy of the model can meet the requirement of manufacturing.