Comparison of two EKF based observers optimized online by both simulated annealing and Big Bang-Big Crunch methods for sensorless estimations in Induction Motor

In this study, it is discussed whether fast converged optimization methods should optimize online the covariance matrices of the Extended Kalman Filter (EKF) based algorithm used in estimating the stator current components, the rotor flux linkage components, the rotor speed and the load torque of Induction Motors(IM) for the sensorless vector control applications. If this scheme is possible, it will provide a solution approach for the sensitivity problem on parameter variations encountered while in estimation process; because the covariance matrices of system noises that are related with these variations will have been tuned online. At this point, speediness becomes restrictive feature on the choice of an appropriate optimization method because of the possible increase on computational load in each estimation sample. For this reason, the Big Bang - Big Crunch, to be the fastest heuristic optimization method, has been employed in determining the covariance matrices in the filter algorithm. Consequently, results from the simulation studies where BB-BC is compared with another heuristic optimization method, so called Simulated Annealing, distinguished that BB-BC is better with respect to convergence speed; however, in terms of reaching optimum solution while under parameter variations, none could perform promising estimation results.