Application of Adaptive Artificial Neural Network Method to Model the Excitation Currents of Synchronous Motors Ramazan Bayindir Ilhami Colak Seref Sagiroglu Hamdi Tolga Kahraman Department of Electrical & Electronic Engineering, Faculty of Technology, Gazi University, Ankara, Turkey, bayindir@gazi.edu.tr Department of Electrical & Electronic Engineering, Faculty of Technology, Gazi University, Ankara, Turkey, icolak@gazi.edu.tr Department of Computer Engineering, Faculty of Engineering, Gazi University, Ankara, Turkey ss@gazi.edu.tr Department of Software Engineering, Faculty of Technology, Karadeniz Technical University, Of, Trabzon, Turkey htolgakahraman@yahoo.com Abstract—In the classic ANN-based approaches, the synchronous motor parameters mostly could be modeled with n-hidden layered networks. It is an important challenge in driver software development is to realize complex mathematical models in real time environments and circuits. This paper presents an Adaptive Artificial Neural Network-based (AANN) method to easily model excitation current of synchronous motors. It has a simple network structure and less processing units (nodes) more than classic ANN. The main purpose of this method are to estimate the excitation current and also to assist designers to model excitation current easily and to develop complex driver software with low degree programming effort while improving the efficiency of classic ANN-based approach. In the adopted approach, the activation functions of nodes in the hidden layers of multilayered feed forward neural network have been determined by using a heuristic method. The experimental results have shown that the proposed method successfully creates single-hidden layered simple networks have less node number than classic ANN-based solutions and achieves the tasks in high estimation accuracies. Keywords: Synchronous Motor; Adaptive Artificial Neural Network; Excitation Current Estimation; Genetic Algorithm I. INTRODUCTION Synchronous motors are generally used in power applications because of their high operating efficiency, reliability, controllable power factor, and relatively low sensitivity to voltage dips. These motors have constant speed and used in mills, refineries, power plants, etc. to drive pumps, fans and other large loads, and also help to assist in power factor correction. In compensation methods using a synchronous motor, excitation current is changed, and absorption of ohmic, inductive or capacitive power from the network is ensured. The amount of active power that a synchronous motor will draw from the network while working unloaded in a reactive power compensation system is as low as the amount needed only for covering the mechanical loss of the motor. A synchronous motor will both generate mechanical energy by means of connection of any load requiring fixed speed to its shaft, and provide compensation without causing any additional cost to the establishment. One of the most important application areas of synchronous motor is the power factor correction for reactive power compensation [1-5]. Many researchers have suggested some estimation methods for improving the compensation with the help of the methods such as proportional plus integral (PI), proportional plus integral plus derivative (PID), pulse width modulation (PWM), fuzzy logic (FL) and artificial neural networks (ANNs) [1, 6-7]. ANN based dynamic reactive power compensator proposed [3-5] provides successful and fast results. In this paper, an Adaptive Artificial Neural Network-based (AANN) method to easily model excitation current of synchronous motors. It has a simple network structure and less processing units (nodes) more than classic ANN. The main purpose of this method are to estimate the excitation current and also to assist designers to model excitation current easily and to develop complex driver software with low degree programming effort while improving the efficiency of classic ANN-based approach. II. THE PROPOSED METHOD A. Adaptive Artificial Neural Network AANN is a modified version of classic ANN. It composed of two units. These are the multilayered 2012 11th International Conference on Machine Learning and Applications 978-0-7695-4913-2/12 $26.00 © 2012 IEEE DOI 10.1109/ICMLA.2012.167 498 2012 11th International Conference on Machine Learning and Applications 978-0-7695-4913-2/12 $26.00 © 2012 IEEE DOI 10.1109/ICMLA.2012.167 498