Research On Nonlinear System Identification Method Based On Neural Network

System identification is an important part of modern control theory.The traditional identification algorithm is usually based on linear,non time-varying and uncertain parameters,and it is difficult to identify general nonlinear time-varying systems.Neural network has been widely applied in the identification of nonlinear systems due to its strong approximation ability,high nonlinearity and strong self-learning ability.The weights of the conventional neural network are usually constant,and the weight of the network is adjusted by the integral learning rate.The time-varying neural network has time-varying weights.When the topological structure of the network is determined,the training of network weights is the key to the successful application of time-varying neural network.In view of the complex nonlinear time-varying dynamic characteristics of nonlinear time-varying systems,the modeling method based on Hopfield neural network theory is studied,and several Hopfield neural network identification algorithms and strategies are proposed in this paper.Neural network system identification is based opn directly learning the input and output data of the system,making the objective function get the minimum value,so as to solve the mathematical description problem of the system,that is,establish the system model.The system identification is divided into two aspects:structure identification and parameter estimation.In this paper,the structure identification based on time varying Hopfield neural network is studied,time varying Hopfield neural network has the strong approximation ability of nonlinear system,in this paper,the mathematical theories are proved the delayed Hopfield neural network approximation delay approximation theorem of constant coefficient diffusion reaction of Hopfield neural network to approximate the reaction diffusion approximation theorem of nonlinear dynamical systems and time-varying reaction diffusion Hopfield neural network approximation time-dependent reaction diffusion approximation theorem of nonlinear dynarmical systems the nonlinear system.These three approximation theorems are universally applicable,which provide theoretical guarantee for solving system modeling with delay phenomena in practical problems,system modeling with diffusion phenomena and system modeling with dynamic process.Real time delay system modeling algorithm and nonlinear reaction diffusion system variable and the iterative learning method and Euler discretization are designed for nonlinear time-varying systems,nonlinear,simulation results verify the feasibility of the specific model of real time modeling algorithm.As the application of this algorithm,Time-varying Hopfield neural network identification algorithm and time-varying delayed Hopfield neural network identification algorithm are applied to wood drying modeling process,and the real-time control model of temperature and humidity in wood drying process is obtained.Numerical simulation results show that the method has good approximation effect.The temperature and humidity control model is a relational model between hot pressing temperature,humidity and moisture content.It provides a necessary condition for studying the intelligent control of wood drying.Secondly,this paper studies the problem of parameter estimation,and takes the internal temperature and heat conduction model of wood in hot pressing process as the research object,and studies the parameter identification of its thermal conductivity and internal heat source.In this paper,the Schauder's fixed point theorem is used to prove the well posedness of the inverse problem of the reaction diffusion equation Neumann.The theorem provides a guarantee for the parameter identification of the inverse problem.Combined with the theory of well posedness of inverse problem,parameter identification of thermal conductivity and internal heat source of internal thermal conduction model in fiberboard hot pressing heat transfer process is carried out.In the identification process parameters,using the central difference scheme and step format method separated by Newton iterative scheme for thermal conductivity equation by using the numerical method,through the simulation experiment of Matlab parameter inversion can be seen close to a good degree and the actual measured value.It helps to better control the heat conduction in the heat transfer process of the fiberboard.In conclusion,neural network model as a nonlinear system identification model is a physical realization of the actual system.The identification principle and method in this paper have general characteristics and can be extended to modeling and identification of other similar nonlinear systems.