Speed Observer Based on ICA Trained Neural Network in DTC drive of IPMSM

In this paper a speed observer based on Imperialist Competitive Algorithm (ICA) trained artificial neural network is presented. The proposed speed observer is used in sensorless Direct Torque Control (DTC) IPMSM drive scheme. A multilayer perception is trained using imperialist competitive algorithm to estimate the rotor speed. Due to artificial neural network characteristics the proposed speed observer works in wide range speed as opposed to previous observers that doesn’t works low speed or high speeds. Since neural network is trained with ICA, optimum weights of neural network are obtained. Simulation results on different conditions show the good performance of proposed speed observer. 1Introduction Permanent magnet synchronous machines (PMSMs) are widely used in many industrial applications, thanks to their compact size, high efficiency, high power density, large torque to inertia ratio and absence of rotor losses [1][4]. These advantages make them good candidates for high-performance applications, such as electric vehicles, wind turbines and robotics [2]. Recently, sensorless PMSM drives have received increasing interest for industrial applications where there are limitations on the use of a position sensor. Furthermore, sensorless control for motor drives reduces susceptibility to noise and vibration, cost, size and maintenance while increasing the overall system’s reliability and robustness when used properly. Over the years, researchers attempted various estimation techniques. The rotor flux linkage estimation method [5], Extended Klaman filters (EKF) [6], [7], Model-based observers [8], [9] and model reference adaptive system (MRAS) observer [10] are the main estimation techniques that presented in literature. Each one of these techniques suffers from some defects such poor precision at low speed, dependency on parameters, and requirement to proper initialization, instability, etc. On another aspect, artificial neural networks (ANNs) and fuzzy logic systems (FLSs) have been known as powerful tools capable of providing robust approximation for mathematically ill-defined systems that may be subjected to structured and unstructured uncertainties. The universal approximation theorem has been the main driving force of such methods as it shows that they are theoretically capable of uniformly approximating any continuous real function to 11-E-ELM-1409 Speed Observer Based on ICA Trained Neural Network in DTC drive of IPMSM 26 International Power System Conference 2 any degree of accuracy. Various ANNs and FLSs have been proposed for speed and position estimation and control of a PMSM, which have led to a satisfactory performance [11], [12]. Direct torque control (DTC) method is selected thanks to it is various advantages such as the elimination of coordinate transformation, lesser parameter dependence, and faster dynamic response in comparison with FOC method [1]. In this paper, we present a neural network based observer to estimate the IPMSM rotor speed. Imperialist competitive algorithm is used to calculate the optimum neural network weights. Simulation results for different condition are presented for validation of proposed speed observer. 2IPMSM MODELING The IPMSM dynamic mathematical model in the d-q axes rotational reference frame can be described by the following equations: v R i L ωL i 1