Study On The Application Of Artificial Neural Network Theory In Active Power Filter

The electric energy is one of the most important energies in today’s society. The powerquality directly affects the development of human society and the comfortable level ofpeople’s life. However, the power quality problems that today’s society is facing is not sooptimistic. The widespread use of kinds of nonlinear and time-varying electronic devicesresults in more and more harmonic pollution to the power system, and this problem causedmany unnecessary troubles for the normal operation of the power system. At the same time,kinds of new instruments and equipments require a higher power quality. Therefore, theharmonics must be removed quickly, and the active power filter appears in this background.The active power filter eliminates the harmonic current in the power system bygenerating a compensating current which is opposite and equal to the harmonic current.There are three key links in the active power filter: harmonic current detection, DC sidevoltage control, instruction current tracking. In this paper, the three key links are improvedby introducing the artificial neural network algorithm. The adaptive linear neural network,neural network recursive integral PI controller and neural network hysteresis-loopcontroller are also designed to optimize the implementations for the three links.About the adaptive algorithm, the realization process and disadvantages of thetraditional LMS algorithm are discussed first in this paper, in consideration of thetraditional LMS algorithm’s disadvantages, a modified LMS algorithm was proposed. Thismodified algorithm improves the quickness and precise accuracy of harmonic detection byswitching the step length between two values.For DC side voltage control, a new control method is proposed. This method combinesthe advantages of PI control, recursive integral algorithm and neural network. Thetraditional PI controller has some disadvantages, for example, the parameters in this kind ofcontroll are fixed and the adaptive capacity to the complex power grid environments is poor.As an alternative to the traditional PI control, it improves these disadvantages of thetraditional PI control, and achieves the purpose of fast response and small overshoot byconstantly revising PI parameters.In the harmonic current tracking link, a neural network hysteresis-loop controller istrained by using the LMBP algorithm and choosing the input and output signals of thehysteresis-loop controller as the mentor signals. Finally, it is used successfully in thecontrol system of the main circuit.