Research Of Improved FFT Application To Harmonic Detection

As the starting point of researching harmonic problems, Harmonic measurement has become one of the first to face the issue of harmonics. The current theory based on Fourier transform harmonic detection methods have been widely used. Detection methods based on FFT to achieve it rather simple, but will produce and fence effect of spectrum leakage. General use of the FFT windowing and interpolation to improve the method, the accuracy improved to meet the requirements, but the process is quite complicated, difficult to analyze the subharmonic and harmonic frequencies similar.For the above problem, we study the least squares method based on FFT and genetic algorithm combined with the improved method to fix the harmonic parameters so as to improve detection accuracy and reduce the purpose of spectral leakage First of all, according to the discrete form of the power of harmonic mathematical model of measured values FFT. Then define the mean square error of least squares, and minimum mean square error. Because the maximum harmonic number N a great impact on the minimum, so we will group the data into training and test group. Training group with minimum mean square error of the data, with test groups of data to calculate the mean square error. If the mean square error has failed to meet a given standard is to adjust the value of N. When the minimum mean square error of a given standard, record the model parameters and obtain the harmonic components. The use of gradient descent to minimize the mean square error of easy to fall into local optima and the speed and accuracy of the algorithm, we introduce the GA for this problem. At the same time the use of genetic algorithm and gradient descent method to minimize mean square error function, when the Bureau of gradient descent into the use of the genetic algorithm optimal approximate the optimal value, while the gradient descent method converges to the global optimum use of the value of the corresponding parameters as test group calculation error model parameters to achieve the purpose to improve the algorithm.Then on the modified FFT with matlab simulation. First of all, given the simulated signal, the simulated signal from the FFT data in and get the initial parameters, using genetic algorithm and least square method by minimizing the mean square error, By changing the length of the training set of data values to determine N When the mean square error of less than a given value of the parameter values obtained Then. Obtained by comparing the FFT parameters and improved FFT the parameters obtained show that the algorithm can obtain high precision harmonic parameters at the same time interval to distinguish the small harmonic components and effective to prevent leakage of the spectrum.