FIR Model Identification Algorithm with Applications
|School||Zhejiang University of Technology|
|Course||Control Theory and Control Engineering|
|Keywords||system identificaiton FIR model identification treatment methods ofabnormality time delay estimation for multivariable systems applied model validation QRdecomposition|
System identification plays a very important role in high performance automationtechnology such as model predictive control nowadays. Products with much more high qualityare required, in current use, to meet the improvements of people’s production and living standard.The productive engineering also depends much more on the automation technology, and theresearch of control theory comes to a certain depth. However, the application of the newtechniques in process industry is still narrow and limited, due to the lack of dynamic modelswhich are supposed to describe the real process accurately and the slot between theory researchand physical application. System identification develops quickly as a branch of automationcontrol in theory study, but lots of application problems needs to do some research. Aim at thephysical application of identification algorithm, this thesis focuses on the following aspects:1.The properties of finite identificaiton response (FIR) model identification algorithm arediscussed.2.Under the situation when the system is subjected to disturbance, lots of“burr”whichoccurs in the results of FIR model identification will destroy the application of the estimated FIRmodel. In this case, a punishment item will be added into the criterion function to smooth theestimated model. The merits and drawbacks are discussed about this improvement.3.QR decomposition is applied to improve the accuracy and efficiency of FIR modelidentification algorithm, which can convert the inverse operation to the generalized inverseoperation and keep the matrix in the algorithm from abnormity.4.Two treatment methods in FIR model identification are proposed in accordance withpresence of abnormality. One is called the method of linear interpolation, the other is the methodof identification based on segments of data. The applied strategy of these two methods has beenanalyzed and the guidelines are enumerated. Simulations have verified the feasibility andeffectiveness of these two methods.5. A method of time delay estimation for multivariable systems based on FIR model identification is proposed. There are three steps: the application of FIR model identification, theselection of a reasonable threshold and the time delay estimation. Two different selectionapproaches are given in this thesis. Numerical examples with simulation verify the feasibilityand effectiveness of the proposed method.6.The application problems of FIR model identification algorithm, namely applied modelvalidation, are studied to inspect the quality of the estimated model. The uncertainty area of theestimated model in different grades is discussed for industrial use with different accuracydemands. The influences of the identified data length, disturbances of the system, correlation ofthe inputs and the signal to noise ratio are studied further. All the studies with simulationsprovide inspection of the estimated model.7.Asoftware platform is designed to perform FIR model identification algorithm, togetherwith two treatment methods of abnormality based on C#. The software carries out the datacollection, the friendly man machine interface design and the strong graphics operation etc.,which serves as a link between theory and application.