Data-driven Power Quality Abnormal Disturbance Event Detection Method

Abnormal disturbances in power quality have a negative impact on users,and in severe cases can also cause immeasurable economic losses.In order to reduce the above-mentioned negative effects,it is necessary to quickly and accurately identify the types of abnormal power quality disturbance events after an abnormality occurs in the power system,so as to provide conditions for subsequent cause finding and problem management.The State Grid has built a power quality monitoring system and collected power quality data for the entire grid,which provides a prerequisite for the data-driven power quality abnormal disturbance event detection method.In this context,this paper aims at the problems of low classification accuracy and poor generalization in the existing classification and recognition methods of abnormal power quality disturbance events,and uses power quality data to carry out related research.The main work includes:1.Existing anomaly detection methods for high-dimensional power quality data with time series and large fluctuations have problems such as low accuracy of detection results and slow recognition speed.Therefore,this paper proposes a power quality data based on dynamic density clustering.Anomaly detection method.This method is aimed at high-dimensional power quality data with time series and large fluctuations.Based on the OPTICS algorithm,which is insensitive to initial parameters,the data set is divided into different time slices based on time.The time slice uses the OPTICS algorithm to perform anomaly detection,and the obtained results are marked with data categories and attributes,and then only the data with attribute changes are calculated,which reduces the calculation time overhead.Experiments show that this method improves time efficiency while ensuring the accuracy of anomaly detection results.2.Existing methods for classifying and identifying abnormal power quality disturbance events have problems such as low classification accuracy and poor generalization.Therefore,this paper proposes a data-driven power quality abnormal disturbance event detection method.First,the proposed method based on dynamic density aggregation is used.The similar power quality data anomaly detection method performs anomaly detection,and then extracts the frequent item set from the characteristic indicators of the detected abnormal data to obtain the characteristics of the power quality abnormal disturbance event,and then uses the GRU algorithm to detect the power quality abnormal disturbance event in the power grid.Perform classification and identification.Experiments show that this method enhances generalization while ensuring the accuracy of classification and recognition results.3.Based on the above research results,a data-driven power quality abnormal disturbance event detection prototype system is designed and implemented.The system structure and data structure are designed,and the implementation process of the exception processing module with the function of power quality anomaly detection and abnormal disturbance event classification recognition,and the model learning module with the function of power quality abnormal disturbance event feature extraction and abnormal disturbance event classification model learning are introduced.