Study On Multivariable Process System Identification

The development of Modern Control Theory is the basis of use for Advanced Control. Control Computer, especially Distributed Control System,Fieldbus Control System give a strongly hardware and software platform for the use of Advanced Control and On-line Optimization. The use of Advanced Control need advanced control algorithm, and all the algorithms also need the mathematic models. It means that the model is more accurate, and the control algorithm is more precise, the corresponding production process is smoother. Most of the actual industrial processes are multivariable systems, in order to improve the model accuracy, so the multivariable system identification was studied in the paper. The specific contents are as follows:Firstly, the theoretical foundation of the multivariable subspace identification was studied in the paper, established a unified theoretical framework, and summed up the deterministic and the combined deterministic-stochastic systems identification methods' specific steps to achieve. By the two simulation examples, the methods were verified applicable and accuracy. Then according to the actual situation, a subspace identification method with multiple-data sets is proposed. It was modified by the single data set of subspace state space system identification. It is a very practical value new algorithm, and it's proved that it could make good result in different conditions of the input-output data.Discrete-time identification method identified the continue-time system existed three questions: First, it couldn't get a good mathematic model when the input is continuous signal in a typical time; Second, the model relative error increase when the sampling time decrease; Third, the poles in the z-domain's unit circle were made drift when the sampling time and truncation error were selected incorrectly, then it made the system unstable. In light of this situation, the continuous-time frequency subspace identification method which has nothing to do with the sampling time was studied, in contrast with the discrete-time identification method, the continuous-time frequency subspace identification method could increase about 2 percent precisions.