Optimum Setting of Active Filter Parameters By Using Genetic Algorithms M. GHANDCHI† S. H. HOSSEINI†† S. GHAEMI†† † Islamic Azad University, Ahar Branch ††Faculty of Electrical & Computer Engineering , University of Tabriz, IRAN Abstract: - The aim of this paper is computing of the optimum duty cycle and inductance using Genetic Algorithm to minimize tracking error and operating costs by means of a new current reference. Such a current reference is able to share the responsibilities in power quality deterioration between supply and customer, making the filter to deliver a current with a lower harmonic content, which means a lower power delivered and a lower switching frequency. The use of different control approach may be helpful to achieve a good current control since this approach allows to carry out a control action independent from the electrical parameters. Key-Words: - Active filter, Optimum, Genetic Algorithm, Crossover 1 Introduction Traditionally, passive filters have been used to eliminate line current harmonics and to improve the load power factor. However, in practical applications these passive second-order filters present many disadvantages such as aging and tuning problems, series and parallel resonance, and the requirement to implement one filter per frequency harmonics that needs to be eliminated. In order to overcome these problems, different kinds of active power filters, based on force- commutated devices, have been researched and developed [1], [2]. Particularly, shunt active power filters, using different control strategies, have been widely investigated. They have gradually been recognized as a viable solution to the problems created by high-power nonlinear loads [3], [4]. These filters operate as current sources, connected in parallel with the nonlinear load, and generate the current harmonic components required by the load. In this form the mains only needs to supply the fundamental, avoiding contamination problems along the distribution lines. However, shunt active filters present the disadvantages that are difficult to implement in large scale, the control is complicated [5], [6], and the cost is high. One of the issues most of the researchers agree upon is that non-linear loads should not be considered the only responsible of the detrimental effects related to power systems in the distorted conditions, but the responsibility for the power quality deterioration should be shared between supply and customer [7]. In control applications, calculus_based optimization schemes are often employed to seek for optimal control gains to meet the performance requirements. Genetic algorithms (GAs) are stochastic search techniques deriving inspiration from the principles of natural evolution and population genetics. GAs are versatile optimizers, and require minimal a priori assumptions on the nature of the problem. In contrast with other search strategies, GAs do not require differentiability, continuity, or other restrictive hypotheses on the objective function. Thus, they well lend themselves also to practical performance-oriented control design problems [8]. The aim of this work is to minimize tracking error and operating costs by select of optimum duty cycle and coupling inductance with the GA by means of current reference. 2 Computation current reference Main task in designing an active filter control is to synthesis the current reference waveform for the PWM inverter. A shunt configuration is shown in Fig.1. At the metering section only the voltage and current waveforms, both network conditions and load operating condition are responsible for, can be measured. It is not possible to know anything about the supply_side or the load_side of the same section. Independently from what there is really at the load_side,a load that shows a linear behavior represents an ideal load condition. In fact, if the load under test were linear, the supply system would be the only responsible for the power quality deterioration at the PCC. Therefore, in Proceedings of the 5th WSEAS International Conference on Signal Processing, Istanbul, Turkey, May 27-29, 2006 (pp243-248)