Research On Intelligent Control Of Switched Systems

Switched systems as an important class of hybrid systems consist of a finite number of subsystems and a switching signal which orchestrates the switching among them,with a wide range of their applications.As we all know,real-world systems present to be intrinsically nonlinear,and there exist a lot of uncertainties such as modeling errors,external disturbances,unmodeled dynamics and measuring noises,etc.,and time delay phenomenons in these practical systems,it is hard to obtain their exact mathematic model for these real-world systems,hence it is hard to solve the high performance control problems of the switched systems with uncertain,intrinsically nonlinear and time delay by the traditional control methods.As the requirements for high precision,response speed,fault tolerance,adaptive ability and self-learning ability are increasingly higher in the modern industrial control,intelligent control methods have become an effective means to control uncertain complex nonlinear systems with high performance.Based on the current research findings on switched systems,the priblems on intelligent control for classes of uncertain nonlinearswitched systems and the design method of the switching strategy are deeply investigated in this dissertation.The main research results of this dissertation are stated as follows:1.An adaptive neural dynamic surface tracking control scheme is presented for a class of single input single output uncertain switched nonlinear systems in strict-feedback form with external disturbance under arbitrary switching.In this scheme,dynamic surface control(DSC)technology is introduced into backstepping design approach with common Lyapunov function method.With each step in the backstepping design,the radial basis function neural network is adopted to approximate constructed unknown upper bound smooth function,which eliminates the influence of the switching signal,and with the help of DSC,the explosion of complexity by repeatedly differentiating common virtual controllers is avoided,and the derivatives of filter output variables instead of traditional intermediate variables are taken as the neural network(NN)inputs.As a result,the dimension of NN inputs is reduced.Simultaneously,Young's inequality is used to reduce the number of adjustable parameters of the control scheme.Only one NN is employed in each step.Moreover,it is proved that the proposed scheme is able to guarantee that all the signals in the resulting closed-loop system are semi-globally uniformly ultimately bounded,with tracking error converging to a small neighborhood of zero by appropriately choosing design parameters.Simulation studies are carried out to illustrate the effectiveness of the proposed control.2.An adaptive neural dynamic surface tracking control scheme with prescribed performance is proposed for a class of single input single output uncertain switched nonlinear systems in pure-feedback form with external disturbance under arbitrary switching.By using mean value theorem,the original system in pure-feedback form are transformed into the new one in strict-feedback form with unknown control directions.Then using the output error transformation based on performance functions,the constrained tracking problem of this new system is transformed into the stabilization problem of an equivalent unconstrained one.For this equivalent unconstrained switched system,common Lyapunov function method and dynamic surface control technology are introduced into backstepping design approach.With each step in the backstepping design,the radial basis function neural network is adopted to approximate constructed unknown upper bound smooth function,which eliminates the influence of the switching signal,the Nussbaum function is employed to solve the unknown symbol problem of the unknown control gain,and with the help of DSC,the explosion of complexity by repeatedly differentiating common virtual controllers is avoided,and the derivatives of filter output variables instead of traditional intermediate variables are taken as the neural network(NN)inputs.As a result,the the dimension of NN inputs is reduced.Only one NN is employed in each step.Moreover,it is proved that the proposed scheme is able to guarantee that all the signals in the resulting closed-loop system are semi-globally uniformly ultimately bounded and the tracking error is within the prescribed performance bounds for all times when the initial condition satisfies the predefined performance bounds.Simulation studies are carried out to illustrate the effectiveness of the proposed control.3.A monotonically convergent iterative learning control(ILC)scheme is presented for a class of uncertain discrete-time switched systems with state delay(UDTSDSs).By taking advantage of output error and state information,a hybrid ILC law for a class of UDTSDSs is proposed.After the ILC process is transformed into a 2D system,sufficient conditions in term of linear matrix inequalities(LMIs)are derived by using a multiple Lyapunov-Krasovskii-like functional approach and a quadratic performance function.It is shown that if certain LMIs are met,the tracking error 2-norm converges monotonically to zero along the iteration direction,while the learning gains could be determined directly by solving the LMIs.The simulation results are provided to illustrate the theoretical analysis.4.Base on the classification feature of the switching strategy in multiple models switching control,a fast support vector machine(SVM)modeling method with Gaussian kernel is proposed,which provides a way for designing the switching strategy with SVMs.The theorem that the inter-cluster distance in the feature spaces(ICDF)is a strictly unimodal positive definite function about Gaussian kernel parameter is firstly presented,and a modified golden section algorithm(MGSA)is proposed to search the best Gaussian kernel parameter in small amount of ICDF calculations.Then,the fast modeling method including two stages is presented for SVMs by using MGSA and evolutionary algorithm.In the first stage,MGSA is used to obtain a shrunk value interval for Gaussian kernel parameter.In the second stage,a differential evolutionary algorithm is applied to select the best parameter combination for SVMs in the shrunk interval of kernel parameter obtained by MGSA and a given interval of penalty parameter.The proposed fast method has effectively reduced the calculating amount and increased the speed of SVM modeling.Experiments for benchmark datasets illustrate that the training time of SVM models can significantly shortened by our approach,while the testing accuracy of the trained SVMs is competitive.