Font Size: a A A

Reaction Catalyzed Distribution Parameter Calculation Based Process Modeling Intelligence

Posted on:2015-02-13Degree:MasterType:Thesis
Country:ChinaCandidate:J J YangFull Text:PDF
GTID:2261330428477699Subject:Control Engineering
Abstract/Summary:PDF Full Text Request
This paper studies on a method based on data-driven modeling forchemical catalytic reaction of nonlinear distributed parameter systems (DPS).Firstly, Kernel Principal Component Analysis (KPCA) method is utilized toextract the nonlinear basis functions in dominant space, and the time-spacedecomposition is carried out in terms of these basis functions to obtain theoutputs in time domain. Kernel parameter is optimized by using GeneticAlgorithm (GA) to obtain a proper principal component space. After thetime-space decomposition, a temporal auto-regressive (ARX) model with theexternal input is identified by using the temporal coefficients obtained from thedecomposition along with the excitation input signal, which is gained by theRecursive Least Squares (RLS) algorithm. Secondly, Principal ComponentAnalysis (PCA) method is utilized to extract the basis functions in dominantspace and the ODEs are obtained, and identify the unknown parameters ofdecoupling ARX model. Nonlinear unknown part of the RBF neural network isused for further modeling, at the same time, realizing system. The GA isadopted to optimize RBF center and RLS is used to identify weights. Finally,considering the sensor position which has a significant effect on the parameteridentification precision of the distributed parameter systems, the D-optimumexperimental design and GA are used to optimize the sensor location. Thesimulation results show that the distributed parameter modeling method hassatisfactory ability of processing nonlinear data and modeling accuracy.Location optimization of sensors can contribute to under the premise of the useof the least number of sensors to ensure identification accuracy of modelingparameters.
Keywords/Search Tags:Nonlinear distributed parameter systems, Kernel PrincipalComponent Analysis, Genetic Algorithm, ARX, RBF neural network, Sensorlocation
PDF Full Text Request
Related items