Power Quality Disturbances Recognition Based On Differential Entropy Recursive Analysis And ELM Classification

Quality of electrical energy should meet the requirements of modern production and People’s daily life. Meanwhile it should guarantee social production and people’s basic lives. With the construction of the smart power grid, a variety of intelligent digital equipment access the power grid network, which causes more and more serious disturbance of power quality and have a negative effection on industrial production and people’s life may also cause irreparable damage. Therefore, in order to take corresponding measures to improve power quality, the problem of power quality disturbance detection and classification analysis become a hotspot in the research of the modern scholars widely.Power quality disturbance have various kinds and possess nonlinear and non-stationary characteristics. In addition, each type of disturbance has its characteristics. The effect is uneven using general time-frequency analysis method to deal with different disturbances.Therefore, aimed at the complexity of the disturbance signal, this paper presents a method that feature extraction based on the differential entropy of space reconstruction and the recurrence plots and quantification recurrence analysis. In order to improve the accuracy of classification of disturbance signal, extreme learning machine(ELM) method is introduced to training and classifying the characteristics value.Firstly, the common type and feature extraction method of power quality disturbances are introduced in this paper. Meanwhile, this chapter decomposes the disturbance signal by several time frequency decomposition methods and the resulting of contrast effect is presented.Secondly, the question is not ideal solved by time-frequency decomposition renderings. Considering the non-stationary, nonlinear characteristics of the starting signal,the chaotic phase space reconstruction of differential entropy is introduced to solve the problem. Meanwhile, obtaining optimal delay time and embedding dimension is to avoid the inconsistencies. After getting the parameters of phase space reconstruction, drawing recursion graphs, dynamic characteristics of time series is displayed by visual and figured graph.Then, in order to improve the accuracy of the disturbance classification, characteristic quantity of chaotic time series is introduced. On the basis of recurrence plots, recursive quantitative parameter is selected as recurrence plots quantitative analyze, RQA(Recurrence Quantification Analysis) method is applied to obtaining non-linear characteristic quantity to compose the disturbance signal recognition feature vectors.Classify power signal combined with ELM(Extreme Learning Machine) artificial neural network.Finally, the simulation signal and the real power quality disturbance data from main circuit in Chengde steel plant collected by mutual Inductance Sensor are as the data of this experiment. Experimental results demonstrate the effectiveness of the power quality disturbance recognition based on differential entropy recurrence quantification analysis and ELM classification.