Nonlinear System Identification Based On Structural Adaptive Filtering Method

System identification is the theory and method to study and establish the mathematical model of the system.For the large-scale multi-variable industrial production process(such as refining and distillation process,heavy oil catalytic cracking process,chemical refining process,etc.)with complex structure,many variables and high dimension,it is extremely difficult to establish such a precise model.System identification is based on a large number of observation data to study the corresponding identification algorithm.The mathematical model of the system is obtained through calculation.The modeling accuracy and application effect are evaluated through model validation.In practice,many large-scale calculation problems and optimization problems are nonlinear,which have high complexity and a large amount of calculation.The identification of the nonlinear multi-variable system is an important research direction of system identification.Most of the existing nonlinear system identification methods are parameter identification,which have fixed structure.In view of its high limitation and low identification rate,the concept of adaptive structure is introduced into the identification process,and a subsystem-based structural adaptive filtering algorithm(SSAF)is proposed.SSAF is composed of subsystems.The subsystem has a linear-nonlinear structure,in which the linear part is a 1-order or 2-order infinite impulse response(IIR)digital filter with uncertain parameters,and the nonlinear part is a static nonlinear function.In initialization,subsystems are randomly generated and connected with no feedback branches between subsystems,which ensures the stability of the nonlinear system.An adaptive multiple-elites-guided composite differential evolution algorithm(AMECo DEs)is used to optimize the structural adaptive filtering model.The model is optimized circularly until the optimal or the near-optimal structure and parameters are found.The simulation results show that AMECo DEs outperforms five other state-of-the-art algorithms on comprehensive performance and convergence rate.The experimental results show that the proposed method has good recognition rate and convergence rate for nonlinear test functions and real data sets,which has remarkable ability of model recognition and information extraction for complex nonlinear system.In the structural adaptive model,there are two types of subsystems: the first-order system and the second-order system,which are added to the adaptive model according to a series of instructions.Then,an effective nonlinear filtering model is formed.The structure and parameters of the adaptive filtering model are coded,and then the generated adaptive model is optimized by evolutionary algorithms.First,the experiments are carried out under different super parameters,the results are compared and analyzed,and the optimal super parameters are selected;then the optimal super parameter values are set for six different evolution algorithms.Six algorithms are used to identify the same nonlinear object,so the optimal algorithm is selected.Finally,the optimal algorithm is used to design the structural adaptive model.An actual liquid saturated steam heat exchange system is identified,and the optimal structure and parameters are obtained.Through the analysis and comparison of the experimental results,it is verified that the structural adaptive filtering model performs well in the non-linear test functions and the real data set.Compared with the neural network identification method,the number of parameters used in this model is less,and the identification results are better,which verifies the superiority of the proposed method.The proposed structural adaptive filtering method based on subsystems provides a new idea for related research.Besides,this method is universal and compatible with other evolutionary algorithms,which allows it to be applied to other similar optimization problems.