Genetic Algorithm with Dynamic Selection Based on Quadratic Ranking Applied to Induction Machine Parameters Estimation

This article presents an efficient off-line identification method for induction machines by a real-coded genetic algorithm. Using only the starting current and the corresponding phase voltage, the electrical and mechanical parameters are estimated simultaneously. This is achieved by minimization of the quadratic output error between the current acquired experimentally from the induction machine and the computed one from the adopted model at the same instance. To improve real-coded genetic algorithm performance and avoid a risk of premature convergence, a dynamic selection based on quadratic ranking is proposed for varying the selection pressure across the generation evolution. A comparison of the different real-coded genetic algorithms-the proposed, linear ranking, Roulette wheel and Boltzmann real-coded genetic algorithms-is carried out on two motors' (1.5 and 0.4 kW) parameter estimation. The transient and steady-state computed current using the estimated parameters are best matched to the measured current, proving that the estimated parameters are well suited for these machines. The results obtained show the superiority of the proposed real-coded genetic algorithm versus the other algorithms in terms of computing time and speed convergence.