Parameter Identification Of Nonlinear Systems Based On Improved Bat Algorithms And RBF Neural Networks

Nonlinear system identification is a hot issue in the field of automatic control.Traditional identification methods,such as least squares method and maximum likelihood method,can only deal with parameter identification of linear systems,which is not applicable to nonlinear systems.When using non-linear least squares methods,such as iteration method and random search method,to identify such models,it is easy to fall into "dimension disaster",which leads to poor stability of identification results.Intelligent optimization algorithm provides a good solution to this kind of complex problem model which lacks gradient,continuity and convexity characteristics.Bat algorithm has attracted wide attention of scholars because of its simple structure and strong global search ability.But it has some shortcomings,such as simple population initialization,low convergence accuracy in the early stage,and easy to fall into "premature convergence".In order to improve the search ability of bat algorithm,this paper proposes to improve the search method by introducing Opposite-based leaning into bat algorithm,and designs a dynamic decreasing Cauchy mutation mechanism to help the algorithm jump out of local extremum.The test function proves that the proposed OCBA algorithm has higher convergence rate and convergence accuracy.Traditional parameter identification methods are usually based on white noise or colored noise based on Gauss distribution.Heavy tail distribution noise,which contains large outliers,presents new challenges to existing identification methods.Hammerstein model can be used to describe most practical nonlinear processes in existing nonlinear models.There is no unified identification method for this kind of model under heavy tail noise interference.In this paper,an improved OCBA algorithm is proposed to deal with the parameter identification of this kind of model.The identification problem is successfully transformed into the parameter optimization problem.The simulation results show that the identification method proposed in this paper has a good identification effect.Neural network is an effective means to realize approximation of nonlinear functions.Aiming at the difficulty of parameterization of complex non-linear links in MIMO Hammerstein model,this paper uses RBF neural network to construct non-linear links,and proposes CMBA algorithm to simultaneously optimize the parameters of RBF network and the linear links of Hammerstein model,so as to realize the synchronous identification of the parameters of MIMO Hammerstein model's non-linear links and linear links.It is proved that the proposed CMBA-RBF method is more robust in identifying the parameters of this kind of model.