Adaptive Measurement Technology For Electrical Parameters

Fourier transform is the commonly used method for the analysis of electrical parameters. The paper analyzed the characteristics and laps of the Fourier transform and introduced the neurual network and power spectral density (PSD) estimation in the measurement of the electrical parameters. Two adaptive electrical parameter analysis models, enhanced Adaline neural network and harmonic basis function neural network, were proposed to meet up to the electrical measurement requirement of shorter sample data, higher frequency resolution and faster learning speed.The limitation of Fourier transform in the measurement of electrical parameters is that it exists the spectral leakage and fence effect, its frequency resolution is decided by the sample data length, and it can not be directly applied to non-stationary periodic signal measurement. For the non-stationary periodic signal, the paper presented a searching based synchronizing approach. In the approach, the integer cycles of sequence were truncated from the asynchronous sample data by reverse searching method, and then the electrical parameters were obtained by discrete Fourier transform. For the asynchronous measurment of the stationary periodic signal, there are conflicts between the measurement precision and the computational burden for the windowed interpolating Fourier transform methods such as Hanning, Blackman or Blackman-Harris window. The paper presented an improved Hanning windowed interpolating Fourier transform method which eliminated the spectral leakage on the second hamonic generated by the fundamental component. The improved method promoted the orerall analytical accuracy of the Hanning interpolating Fourier transform with a small amount of additional computation.Adaline neural network measures electrical parameters through the principle of adaptive filtering, without prior sample training of the neural network weights. But Adaline requests the signal frequency known in advance, otherwise small frequency deviation may generate large analytical error. The paper proposed improved enhanced Adaline neural network and harmonic basis function (HBF) neural network models, which treated the fundamental frequency as the weight to be determined, so as to estimate the frequency, amplitudes and phases at the same time. The momentum and the delayed frequency adjustment were adopted in the learning algorithms to promote the convergence speed. The convergence conditions of the adaptive algorithms were analyzed, the impacts of the learning rates and the momentum factors on the algrithm convergence performances were discussed, and their optimal ranges of choices were presented. The simulation examples demonstrated that the enhanced Adaline neural network and the HBF neural network approaches achieved high precision and rapid convergence.An interhamonic analysis method of combining modern PSD emstimation with the adaptive neural network was presented to overcome the disadvantage of Fourier transform of low frequency resolution. In the supposed method, AR Burg algorithm or MUSIC algorithm was used to estimate the numbers and the pre-estimated frequencies of harmonics and interharmonics, and then all the frequencies, amplitudes and phases were obtained by the enhanced Adaline or HBF neural network. The simulation results showed that the proposed method had the advantages of high frequency resolution, good analytical accuracy and fast convergence speed, and was applicable for interhamonic analysis of asynchronous sampling and short data length.Finally, the electrical comsuptions of computers and several home appliances were sampled through an electrical data acquisition system, and their electrical parameters were analyzed by the adaptive neural network method. The analysis results further verified the validity and applicability of the proposed adaptive measurement methods for electrical parameters, as well as obtained the electricity comsuption data and characteristics of computers and several home appliances.