A Systematic Approach to Estimate Non-Linear System Parameters using Particle Swarm Optimization and Bond Graph Methods

Abstract

This paper presents an application of particle swarm optimization (PSO) methods to predict Bond Graph parameters. In a real-world application, the parameters within the bond graph model can be unknown due to various reasons, such as weariness and unknown disturbance. Hence, the parameters can be identified using metaheuristic methods, such as PSO. Therefore, several PSO variants are applied and investigated to predict the unknown parameters for two nonlinear systems. The modeled systems are a single tank case representing a simple incompressible fluid system and a passive actuator in turbocharger representing interconnected multi-domain system behaving highly nonlinear considering the pulsating flow exhaust gas pressure as the device's input. The investigated PSO systems are the classic, linearly decreasing inertia weight, and constriction PSO. The objective function is formulated to estimate the equivalent pipe area in the tank system and two parameters in the passive actuator system. After the optimization process, the predicted response has an agreement with the experimental results. The linearly decrease inertia weight PSO has shown comparable performance with constriction PSO with root mean square error up to 0.7 for a tank system and 2.5 for the passive actuator. The applied PSO shows satisfying results and shows its capability to predict unknown variables for both cases.