Novel swarm intelligence-based multiple models approaches for parameter estimation: Application to a railway vehicle system with traction

Abstract

Multiple-models approach is one technique for estimating the parameter values of the system of interest. This approach involves creating multiple models of the physical system, each operating with a different parameter vector but using the same initial conditions. However, noise in measurements obtained from these systems and model discrepancies often degrade estimation performance. This study introduces two novel swarm intelligence-based approaches for the multiple-model parameter estimation technique, specifically designed for systems with noisy measurements and model discrepancies. These new approaches are inspired by the bare bones particle swarm and grey wolf optimization techniques. To evaluate their performance, the approaches are applied to a tram wheel test stand characterized by noisy measurements and model discrepancies, where the normal load on the wheel is estimated. The results demonstrate that the proposed approaches eliminate the need for velocity clamping found in the previously proposed technique, and the bare bones particle swarm optimization-inspired approach improves estimation accuracy compared to both the particle swarm optimization- and grey wolf optimization-inspired methods.