Font Size: a A A

Research On The Method Of Multi-modeling Soft Sensor Based On Double Layer Intelligent Structure

Posted on:2009-03-31Degree:MasterType:Thesis
Country:ChinaCandidate:L S ShiFull Text:PDF
GTID:2178360245956843Subject:Control theory and control engineering
Abstract/Summary:PDF Full Text Request
With the development of the economy and the increase of the information for large petrochemical enterprises, as an efective method of realizing high-quality, stable and balanced production, the advanced control gets wider application.Soft sensor technology is an effective method to solve the problem of real-time measurement of the important production parameters such as the quality. It also gets extensive attention and recognition owing to the fact that it can improve benefit of production and guarantee the quality of production.It is difficult for single model to describe global properties of complex system, and multi-point of complex systems in work is taken into account, so a multi-modeling soft sensor based on double layers intelligent structure is proposed in this paper, in which fuzzy c-means clustering is classification layer, and the RBF neural network and and least square support vector machine (LS-SVM) are modeling layers. The degrees of membership are used for conbining the output of submodels to obtain the finial result which is the measured value of estimated parameters.In modeling the multi-model soft measurement, parameters selection of least square support vector machine is a difficult problem, and least square support vector machine is not scarce, so a model of PSO-VB-LSSVM is proposed in this paper, in which the model parameters are selected by PSO algorithm, and the support vectors are selected by vector base learning, which makes the support vectors sparse.At last, according to the different industrial background, the multi-modeling soft sensor based on double layers intelligent structure is used to estimate ethylene concentration at the bottom of ethylene rectifying conlumn, and through simulation comparison of several modeling methods, the multi-modeling soft sensor based on double layers intelligent structure in which the PSO-VB-LSSVM is used has better generalization result and forecast accuracy than other methods; least square support vector machine and its combination of particle swarm optimization and vector-based selection criterion are used to estimate research octane number (RON) in the system of gasoline blending, through the simulation of comparison, the methord of PSO-VB-LSSVM has the character of small error and strong generalization ability.
Keywords/Search Tags:soft sensor, double layer intelligent structure, multi-modeling, least square support vector machine, particle swarm optimization, vector base, generalization ability
PDF Full Text Request
Related items