Identification of a Class of Discrete Event Systems by Neural Networks - Sparse Realization -

In analysis and design of a system, the first goal is to obtain an appropriate model of the system. Because of the complex dynamics of discrete event systems (DESs), it is very difficult to obtain a model of unknown DESs from given input and output data. This paper discusses an identification and realization method of a class of DESs by neural networks. We consider a class of DESs which is modeled by using finite state automata. In identification and realization of systems by using neural networks, it is essentially important to develop a suitable architecture of neural networks. We already proposed two neural network architectures: one is a class of recurrent neural networks and the other is a class of recurrent high-order neural networks, which are capable of representing FSA with the network size being smaller than the existing neural network models. In this paper we present an identification method of DESs, which makes it possible to obtain sparse realization, that is, to obtain networks with simpler structure. It is shown through numerical experiments that presented method makes it possible to obtain simpler neural networks which can exactly simulate target DESs.