Shigenobu Yamawaki

Expanded Neural Networks in System Identification

The neural networks are recognized to possess the fault tolerance and learning capability. The neural networks are also used in the identification of nonlinear systems. However in the system identification it is important to whiten a color noise using the noise model. In this paper we propose an expanded neural network in which a noise model is incorporated into the output layer of the neural network. We have developed the learning algorithm converged more quickly than a classical back-propagation algorithm. The proposed algorithm estimates the parameter of the expanded neural network using the least-squares method, and estimates threshold by the fundamental error back-propagation method.

Robust System Identification Using Neural Networks

The robust system identification method using the neural network is developed based on the canonical variate analysis (CVA). The main contribution of this algorithm is using CVA to obtain the k-step optimal prediction value. Therefore, the method to obtain the comparatively accurate estimate is introduced without iteration calculations. We show that this algorithm can be applied to successfully identify the nonlinear system in the presence of comparatively loud noise. Results from several simulation studies have been included to the effectiveness of this method.

The study of the robust learning algorithm for neural networks

In this paper, we propose the robust learning algorithm for neural networks. The suggested algorithm is obtaining the expanded Kalman filter in the Krein space. We show that this algorithm can be applied to identify the nonlinear system in the presence of the observed noise and system noise.

The Study of the Effectiveness Using the Expanded Neural Network in System Identification

We have proposed the expanded neural network which the noise model has incorporated into the output layer of the neural network. The expanded neural network is able to apply to the output error model for the identification of a nonlinear system. In this paper, we consider whether the expanded neural network is able to apply effectively to estimate the nonlinear system that has a system noise. It is shown that the estimated accuracy is improved with the included noise model also in this case from the simulation.