Steady-state performance constraints for dynamical models based on RBF networks

This paper is concerned with building RBF dynamical models. The work presents a procedure by which a dynamical model is constrained using information about the system steady-state behavior. Numerical results with simulated and measured data show that the constrained RBF models have a much improved steady-state. For noise-free data such improvement happens with no obvious degradation in dynamical performance which only happens when the steady-state behavior is heavily weighed. For noisy data, however, the constrained models are superior both in steady-state and dynamically. The paper also discusses other situations in which the use of steady-state constraints turn out to be advantageous.