Power Quality Analysis Based On Detrended Multifractal And Improved Extreme Learning Machine

Power quality refers to the quality of power supplied by the power system to users.In order to meet people's demand for high-quality power in their daily work and life,the analysis and identification of power quality disturbance signals is particularly important.Based on the analysis of the current research status of power quality disturbance detection,this paper presents a Multifractal detrended fluctuation analysis based on the decomposition of power signals,the construction of eigenvector matrix and classification and recognition.,MFDFA and Monarch butterfly optimization(MBO)improve the power quality analysis method of Extreme Learning Machine(ELM).Firstly,the principles and characteristics of several common time-frequency domain analysis methods are analyzed,and their limitations are pointed out.A feature extraction method for power quality disturbance signal based on MFDFA is presented.The MFDFA method excavates the fractal characteristics of the disturbance signal from the geometric level,divides the signal into non-repeating small intervals,uses the least squares method to fit the trend of each small interval,determines the fluctuation function of the interval,and obtains the fractal parameters of the signal.The multifractal characteristics of power quality disturbance signals are demonstrated by the generalized Hurst theory and the multifractal spectrum analysis of signals.Secondly,the obtained fractal parameters are selected,and three fractal parameters that can characterize the signal characteristics are selected.To further highlight the characteristics of different types of disturbance signals,the sample entropy of the multifractal spectrum and the energy entropy of the signal itself are calculated to provide the eigenvectors for the classification and recognition of disturbance signals of power quality.Three common signal feature extraction methods are selected to compare with MFDFA.Experiments show that this method performs better in noise resistance and classification accuracy.Next,by analyzing the principle,advantages and disadvantages of ELM,an improved classification model based on Monarch butterfly optimization(MBO)algorithm is presented.MBO algorithm is used to optimize the weights and thresholds generated randomly in the ELM,to determine the optimal network parameters,and to determine the number of iterations,the excitation function and the number of hidden elements in the MBO-ELM model through signal simulation experiments.Finally,the methods proposed in this paper are experimentally verified and applied.The validity of the method presented in this paper is verified by using 13 simulated signals of single and compound disturbance signals of power quality and the measured signals of power quality signals of a steel plant in Chengde.At the same time,MBO-ELM model is compared with several common classifier models,which proves that this method achieves better recognition results.