Recognition Of Power Quality Composite Disturbances With Unknown Types Based On Multi Label Classifier

With the application of a large number of power electronic devices,reactive power compensation devices and the integration of distributed new energy such as wind energy and solar energy into the distribution network,power quality disturbances occur frequently and increase continuously.Power quality disturbance events seriously affect the safe operation of power grid equipment and lines,and also lead to malfunction of relay protection equipment,power measurement error,interference with communication and other problems,which are directly related to residents’ life and social production.The number of power quality monitoring points in distribution network is huge,the transmission cost of massive power quality disturbance data is high,and the power quality disturbance information is easy to be lost in the process of signal transmission,which puts forward higher efficiency and accuracy requirements for the transmission of power quality disturbance signal.In the increasingly complex power system operation environment,the possibility of power quality composite disturbance in the actual power system is greater.The causes of composite disturbance are complex,and the time-frequency domain characteristics are irregular.It is difficult to include all types of composite disturbance in the training samples,and it is impossible to accurately identify the unknown types of composite disturbance without training samples.This research completes the compression and reconstruction of power quality composite disturbance signal,and realizes the efficient and accurate identification of power quality composite disturbance based on efficient time domain feature extraction and multi label light gradient boosting machine.Aiming at the problems of high transmission cost of massive power quality disturbance data and easy loss of disturbance information in transmission,a signal compression and reconstruction method based on compressed sensing is adopted.Firstly,the Fourier transform basis is used to sparse represent the power quality disturbance signal,so as to improve the sampling speed and reduce the storage space;Then,the deterministic structured cyclic sparse measurement matrix is used to observe the power quality disturbance signal;Finally,the regularized adaptive matching pursuit algorithm is used to reconstruct the power quality disturbance signal.The traditional power quality disturbance identification method is inefficient,and it is easy to mistakenly identify the unknown type composite disturbance as the known type disturbance.A power quality disturbance identification method based on efficient time domain feature extraction and multi label light gradient boosting machine is proposed.Firstly,the original disturbance signal is segmented by using a unified time domain scale,and 20 time domain features are extracted;Then,the light gradient boosting machine sub classifier corresponding to a single disturbance label is sorted based on split feature importance;Then,based on the classification accuracy of each sub classifier,the feature selection is carried out to determine the optimal feature subset matched with each sub classifier;Finally,according to the optimal sub classifier,the single disturbance component contained in the signal is identified,and a multi label light gradient boosting machine with multiple sub classifiers is constructed.Finally,the recognition of composite disturbance of power quality is realized.This research realizes the efficient and accurate transmission of power quality disturbance data.Simulation experiments show that the new method has high reconstruction accuracy,still maintains good reconstruction effect in the scenario of high compression ratio,and completely retains the disturbance information in the original signal.The simulation and measured data experiments show that the new method has high recognition accuracy and strong anti noise ability,and can effectively identify the unknown type composite disturbance data without training samples,which can meet the requirements of efficiency and accuracy of sea volume and power quality disturbance event identification in practical industrial applications.If the relevant results of this study are applied to power quality monitoring equipment,it will effectively reduce the data transmission cost of power quality disturbance,improve the accuracy and efficiency of power quality disturbance identification,help to ensure the power consumption of residents and social production,and help the high-quality and healthy development of power grid under the background of double ‘carbon’.