NeSS: a Simulation Environment for Behavioral Design of Neural Networks for Prediction and Control

Abstract

The NeSS (Neural Systems Simulator) environment is presented in this paper: it is a exible software package which has been developed to support, analyze and model dynamic non-linear systems for prediction, system identification and control applications, by providing both classical and innovative approaches within a exible and high-level framework. The behavior of each system is easily defined in a graphic way by interconnecting parametrized atomic objects (e.g., algebraic functions and neural networks), whose behaviors can be either predefined or identified by means of a learning procedure. Neural networks play a relevant role in NeSS: rich and easily expandable libraries are given which support different neural structures and learning algorithms.