Multi-Variable Process Modeling Using Statistical Process Data

Modern industry has the features of large scale, high complexity, multi variables and operation under closed loop control. Reasonable modeling of the multivariable production is the key of knowing about and supervising production. With the application of new automation equipments and communication technology, especially the widespread application of DCS system, there is a great quantity of information from the data of the SCADA system, regarding production and quality. Building mathematical model from these huge data to improve control performance attracts both academy and industry.The thesis is mainly about multivariable process modelling. After reviewing the basic means of multivariable modelling from statistical data, two statistical models are set up for CO₂ refinery and wastewater treatment processes respectively. The thesis is arranged as follows:Chapter One introduces the basic theory and approaches of the multivariable process modelling, which are important to control a multivariable process.In Chapter Two, the basic statistical modelling approaches ofmultivariable processes are introduced in detail, such as multi-component linear recursion, Principal Component Recursion (PCR), and Partial Least Squares (PLS), artificial neural network approach.Chapter Three proposes a new modeling approach, which is composed of back mapping PCA and radial basis function neural network, to overcome the problem of data redundancy and nonlinearity. A CO₂ refinery process modelling illustration shows its better precision than the principal component recursion modelling method.After the description of the SBR wastewater treatment processing characteristics and the key water quality parameters in Chapter Four, the Elman dynamic network is introduced in Chapter Five to model such a dynamic non-linear process. The dynamic predictive models of COD etc. are established, the simulation results show that Elman network is a reliable, fast predicting approach with high precision.In Chapter Six, a conclusion of the thesis is made, and the further development areas of the multivariable process modelling are discussed.