Adaptive filtering and signal reconstruction methods for high-precision electrical parameter measurement systems

Abstract

Aiming at the problem that the complex electromagnetic environment leads to the decline of electrical parameter detection accuracy and the lag of transient response under the background of new power system, a highly robust signal processing technology scheme is studied. Firstly, a mathematical model of mixed interference including additive white noise and nonlinear distortion is constructed, and an improved Normalized Least Mean Square (NLMS) algorithm with variable step size based on mutual information gain criterion is proposed. The iterative process of weights is dynamically optimized by the statistical correlation between quantized input and error. On this basis, sparse Bayesian learning strategy is introduced to construct wavelet packet-Fourier joint basis to complete signal undersampling reconstruction, which fundamentally eliminates spectrum leakage and fence effect. The experimental verification shows that the steady-state mean square error of this scheme is stable at -45dB, the improvement of signal-to-noise ratio is 18.6dB, the absolute error of voltage measurement is only 0.85mV, the maximum frequency deviation is 0.047Hz, and the Total Harmonic Distortion (THD) measurement error is as low as 0.12%. All the indexes are better than the traditional reference scheme by about 40%. This research improves the system's parameter tracking ability and measurement accuracy under strong interference and dynamic load, solves the bottleneck of traditional filter phase distortion, and meets the strict requirements of high-precision relay protection and power quality monitoring.