Research On Several Problems In System Identification

Nonlinear system, closed loop system, multi variable system and structure identification are hot spots and difficulties. As for nonlinear system, there will be greater error if the linear model fits the complex nonlinear system. Moreover, one dynamic model which can describe the actual process of the system accurately is difficult to find; for closed loop system, due to existence of feedback effect, the information contents of input/output decrease; in addition, as input influences output noise, the undecipherable problem will occur to the system. As far as multi variable system is concerned, it possesses many variables, complex structure and strong coupling, there always are greater error in traditional identification approach, or this method can’t identify the error. The commonly used system structure identification methods often aim at specific system, which doesn’t belong to universal type. This thesis researches problems aforementioned; several problems included are as follows:1. The traditional identification algorithm, affected by outside colored noise, is difficult to gain consistent unbiased estimate of the model parameters. This thesis comes up with single variable linear and nonlinear system identification approach based on HPSO. This method changes identification problem of nonlinear system into optimization problem in parameter space, then HPSO is applied to searching the whole parameter space, the optimized estimation of system parameters can be gained ultimately. The concept of health degree is introduced by HPSO algorithm and is structured one particle self-cure model. The health situation of particle is monitored dynamically, the sick particle should be mutated specifically, and then purpose of phenomenon self-cure can be achieved. As for single variable linear and nonlinear system identification, the identification approach is capable of obtaining accurate identification results.2. Compared with parameters of identification system model, its structure is much more difficult, to realize simultaneous identification on parameters and structure of identification system model is the most difficult. Therefore, this thesis puts forward one new method called unified identification once, which is based on single variable and multi variable system structure and parameters, it can achieve the purpose of making simultaneous identification on structure and parameters of system model. This method introduces cooperated optimization algorithm to identify system parameters, which possesses global search competence accurately and efficiently as well as partial search capability quickly and precisely, then the estimation on model parameters can be made accurately. The strategy of meta-model fitting is applied in identification of system structure, through permutation and combination of meta-model and input variables, a great many submodels are constructed, What we need to do is to select the submodels fitting the actual system most from such submodels. Then, the structure identification can be changed into problem of modeling optimization, which provides the identification system structure with new thought.3. Swarm intelligence optimization algorithm is one random search optimization algorithm, as the swarm intelligence optimization algorithm is applied to system identification, the initialization process of model parameters is one random process, which leads the algorithm to make blind search in such process. Thus, this thesis comes up with one new approach (reduce the search scope of the model parameters) and it is applied to the identification of closed loop system. This approach improves traditional genetic algorithm, approximate neighborhood range of initial value of the model parameters is estimated by this algorithm. The parameters initial value of the system model is minified to one smaller section, and then the algorithm will not search blindly and identify the model parameters on purpose. Thus, precision and rate of convergence of the algorithm is enhanced, it can reduce many adverse effect due to randomness of algorithm effectively.4. It puts forwards higher requirements on signals of traditional closed loop identification algorithm, as far as high-quality signals are always hard to come by on actual industry field, this thesis comes up with one closed loop identification approach, which is one time delay multi variable system based on random testing signals. This method changes closed loop system into open loop system equivalently, which is based on equivalent open loop conversion at first. It is assumed that r(t) and y(t) are input and output of the system, then transfer function of the closed loop system is Y /R= GcGs(1+GcGs). Based on thought of open loop identification, this thesis takes e= r-y as input of the system, thus the system’s transfer function becomes:Y/E-GcGs, the closed loop system is equally changed into open loop system; identification complexity of the closed loop system is reduced greatly. At the same time, this approach extends the scope of entry signal, which is not limited to step signal, lowers the identification algorithm’s requirement on system identification environment.