Research On Adaptive Harmonic Current Detection Method For Active Power Filter

As is known, harmonic has went into electric network because of plenty of modern electronics coming out. Harmonic can cause serious damage to electric equipment like shorten electric equipment service life, burn out electric equipment and cause electric equipment abnormal operation. Harmonic should be eliminated due to its damage to people’s life. Eliminating harmonic has become a research emphasis for scientists from all countries. And the test for harmonic plays an important part in eliminating harmonic. This paper mainly studies the test of harmonic based on neural network.First, the study construct linear neural network harmonic test system by using the adaptive noise canceling technology. The basic theory of using adaptive noise canceling to test harmonic current is:take load-current as systematic desired signals, of load-current, take harmonic component as inputting noise signals, input reference signals to filters, use the adaptability of adaptive algorithm, continuously adjust filters’ weight vector, minus outputting signals when the algorithm converges to a steady state and the obtained signals approach to fundamental current on phase and amplitude, obtain estimated harmonic current and all steps above make harmonic current test work.Then, put forward two stage cascade linear neural network structure according to the contradictories for choosing linear neural network learning factor, apply the structure to harmonic test, the theory of two stage cascade linear neural network test is:the first stage linear neural network occupies a larger learning rate choosing range, while the second stage linear neural network occupies a smaller one, adjust weight parameters in order to make the first stage linear neural network take a leading position during training initial stage, weight parameters will adjust automatically when training comes to a steady state which leads the second stage linear neural network to a leading position instead. All steps above ensure high convergence rates and minor errors in two stage cascade linear neural network harmonic test.At last, put forward mixed BP-RBE neural network according to BP neural network harmonic test and RBE neural network harmonic test defects. Discuss the feasibility of BP-RBE neural network harmonic test and make simulation analysis by using MATLAB software.