Analysis o harmonics losses in power systems including non-linear loads with artificial neural network and estimating filter costs

In the electrical power system, energy is supplied at a single and constant frequency, and at specified voltage levels of constant magnitudes. However, none of these conditions are fulfilled in practice. The problem of voltage and frequency deviations, and the means of keeping them under control, is the subject matter of modern power system analysis. In fact the problem of the waveform distortion (especially due to harmonics) is not a new phenomenon. But the recent growing concern for this problem (for the last twenty years) results from the increasing numbers, applications and power rating of the equipment and devices which have non-linear voltage-current characteristics. In the first chapter of this thesis, after general introduction previous studies are explained. In the second chapter, artificial neural networks (ANN), training algorithms, learning rules are studied. In this chapter applications in power systems of ANN's are also presented. In the third chapter general mathematical expressions and concepts for harmonics, harmonic sources and effects of them are given. In the 4th chapter, the elimination of harmonics and filter (active and passive) design are explained. In the 5th chapter, a procedure for additional harmonics losses and cost functions for both passive and active filters are derived and harmonic-cost analysis is performed. In the numerical illustration an 8-bus power system including non-linear loads is considered. After fundamental and harmonic load flow is performed using a simulation program, firstly harmonic losses are computed and then these losses -for changing harmonic spectrum situations- are found by ANN. Secondly, using proposed expressions (in Chapter 5) and obtained computer results the cost of the additional harmonic losses and filter for low- voltage are computed in TL (Turkish Lira). The economic comparison of passive and active filters is given in table. Finally, the overall conclusions are discussed. Simulation results based on DIGSilent Power System Analysis Software. C++ software for ANN is given the Appendix. Keywords: Filter cost, harmonic losses, non-linear load, ANN, training algorithm