Robust Identification Of Nonlinear Block-oriented Systems Under Complex Noise Conditions

In system identification,it is an important step to establish the model to be identified.The nonlinear block-oriented systems have a simple structure,which are composed of different static nonlinear blocks and dynamic linear blocks.Its advantage is to separate the dynamic and static phase of system.It is able to describe the actual nonlinear systems.Moreover,the nonlinear and time-varying characteristics of the actual process are able to be more comprehensively described by the design of time-varying nonlinear block-oriented system and switched nonlinear block-oriented system.It is difficult to identify the nonlinear block-oriented systems with the input or output nonlinearity,and the identification results have bias in the complex noise conditions.For Gaussian white or colored noise,the identification algorithm has certain robustness.However,when the pulse interference exists,a huge error is produced by the identification algorithm at current time.The bias will affect the next identification results.Therefore,it is necessary to ensure the identification algorithm of nonlinear block-oriented systems is robust.For time-varying nonlinear block-oriented system and switched nonlinear block-oriented system,not only the time-varying characteristics and switched mechanism of the system,but also the robustness of the algorithm needs to be considered to obtain accurate robust parameter estimation.For the simple nonlinear block-oriented system,time-varying nonlinear block-oriented system and switched nonlinear block-oriented system,there are three proposed robust identification algorithms under complex noise conditions in this paper.For the simple nonlinear block-oriented system,the identification problem is transformed into the minimum optimization problem,and the intelligent identification scheme is obtained by combining particle swarm optimization(PSO)and K-means clustering.PSO is used to solve the identification problem of the simple block-oriented system under non-Gaussian noise,which is combined with K-means clustering to prevent the algorithm from falling into local optimum effectively and get accurate robust parameter estimation quickly.For the time-varying nonlinear block-oriented system,a recursive identification algorithm based on multiple iterations is proposed.In the recursive identification,the idea of multi iteration is introduced to make the information vector contain the information of multiple times so as to weaken the impact of outliers,and a suitable estimator is designed to estimate the statistical characteristics of unknown noise to improve the robustness of the algorithm.The damping coefficient is added to the multi information vector to enhance the proportion of current information,and the appropriate forgetting factor is designed to track the parameter changes and master the time-varying characteristics of the system.For the switched nonlinear block-oriented system,a switched detection scheme based on two recursive identifiers is proposed,and the least absolute criterion is used as the performance index of the recursive identification to enhance the robustness of the algorithm.The identifier with long iterations is responsible for resisting the interference of outliers,and the identifier with short iterations is responsible for tracking the changes of the process.The difference between the results of the two recognizers is analyzed,and the switch mechanism of the slow switched system is obtained.Finally,the convergence of the proposed recursive identification algorithm is proved in theory.Simulation examples indicate that the proposed three algorithms have the effectiveness and robustness.