Recurrent Neural Network Optimization Algorithm And Application In Four-tank Benchmark

In the late 19 th century,Hopfield and Tank proposed an artificial neural network to solve the optimization problem.It opened up a new research approach for the development of artificial neural network.Since then,it has a heated discussion in the field of neural network optimization’s research and application.Nowadays,optimization problems are widely used in function approximation,signal processing,image storage,parameter estimation,mechanical control and so on.Due to the computation time being closely related to the complexity of the algorithm and the structure of problem,the traditional numerical optimization algorithm is of poor performance.However,the artificial neural network is adaptive and parallel,it has achieved good results for solving the optimization problem in recent years.Therefore,it is of great significance to study the stability of neural network in solving the optimization problem both in theory and in practical engineering application.Based on the above problems,this paper further studies how to optimize the application of four-tank with neural network based on convex optimization theory and global exponential stability theory.Specific research content and innovation are as follows,1.Using the model predictive control strategy,four-tank benchmark problem is precisely described as a constrained quadratic programming problem.Based on the Lagrangian multiplier method,the projection method and the saddle point theorem,a class of discrete recurrent neural network is proposed to solve the four-tank benchmark problem.Compared with the other four-tank benchmark optimization problems,the proposed discrete recurrent neural network has the advantages of simple structure and easy hardware implementation.In addition,we also design a circuit model which can show its characteristics better for the proposed recursive neural network.The neural network optimization algorithm is verified by the theory,which can eliminate the state error and control error of the four-tank benchmark problem effectively and make it globally exponentially stable.2.We use KKT condition to establish an one-layer neural network with discrete activation function.This one-layer neural network can solve four-tank benchmark optimized problem.We propose a single-layer recurrent neural network optimization algorithm to eliminate the exact state of the four-tank problem and control error,so that the system can achieve stability.At the same time,the theoretical results show that the presented one-layer recurrent neural network is stable globally for solving four-tank benchmark optimization problem.