Reserch On Neural Network Learning Algorithm Based On Ellipsoidal Set-membership Estimation Theory

Based on the existing results of set-membership estimation theory and its application, the further research on set-membership estimation and its application are carried out in this thesis. And the issues of state estimation of unknown but bounded and learning algorithm of neural network are discussed in details.Firstly, based on analyzing the linear system state bounding estimation under ellipsoidal set description, a recursive state bounding estimation algorithm on condition that system and measurement noise are unknown but bounded is proposed. The simulation example shows that the performance of the algorithm is close to Kalman filtering method. Besides, with analyzing the nonlinear system state bounding estimation, an extended optimal bounding ellipsoid state estimation algorithm is presented is gained, Just like the EKF, the algorithm linearizes the system state equations. the algorithm is characteristic of simple description and effective estimationThen, neural network and learning algorithm of the BP network structure of are studied, and ellipsoidal set-membership estimation theory is applied to BP neural network, and the BP neural network learning algorithm is given on the basis of ellipsoidal set-membership theory. The algorithm is applied outside the optimal set bounding ellipsoid algorithm to complete the training of network weights. Simulation results show that the algorithm has faster learning speed than the traditional learning algorithm.Finally, neural network and learning algorithm of the RBF network structure of are analyzed, and ellipsoidal set-membership estimation theory is applied to RBF neural network, and the RBF neural network learning algorithm is proposed on the basis of ellipsoidal set-membership theory. The algorithm trained the parameter which used by all of the Radial Basis Function to achieve the optimal solution. Simulation shows the effectiveness of the algorithm.