Research On Cleaning Method Of Power Quality Data Based On Correlation Analysis

In recent years,various power quality interference sources such as new energy power generation,high-speed railways,electric vehicles,and frequency converters are widely distributed in the power grid,and their random volatility and nonlinear characteristics are significant,resulting in deterioration of the power quality of the utility grid.The power grid harmonic monitoring and analysis system under construction collects the power quality data of the whole network in real time,making data-driven power quality problem analysis and decision making possible.However,the number of power grid monitoring terminals is large,the storage system is complex,and the physical environment is bad.The data will be missing,abnormal,and other quality problems.These data problems will lead to erroneous analysis results and affect the power quality management decision-making effect.Aiming at the above problems,according to the characteristics of strong timing of power quality monitoring data,high data acquisition frequency and many types of monitoring indicators,a multi-indicator correlation analysis method is proposed to study the missing data filling and abnormal data detection techniques in power quality monitoring data.The main tasks carried out include:1.Filling in the missing value of power quality monitoring data as the research goal.According to the large number of monitoring indicators and the strong correlation of measurement indicators,a missing value filling method based on the correlation of measurement indicators is proposed.The method first introduces the normalized mutual information method to make up for the deficiency of the traditional Pearson correlation coefficient,calculate the correlation of the measurement indicators,and obtain the strong correlation index.Then,using the local weighted regression algorithm,the regression analysis of the obtained indicators is carried out to establish a regression mathematical model between the strong correlation indicators,and finally the model is used for filling.The experimental results show that the missing data filling method based on the correlation of measurement indicators has higher accuracy.The absolute error percentage is used as the evaluation index,and the error rate is 8%.The effect is good and can meet the requirements of power quality analysis.2.The abnormal data detection of power quality monitoring data is taken as the research goal,and the control chart based method is designed to detect whether there is abnormal data in the data.According to the strong correlation of monitoring data,the correlation based on the correlation of measurement indicators is proposed.Detection method;for the measurement index without strong correlation,according to the timing characteristics of the monitoring data,the ARIMA-based anomaly data detection method is designed to screen the abnormal data.The experimental results show that the control pattern anomaly detection method is faster when the steady-state control map is formed,and the detection time of 2555 indicators is about 67 seconds.The detection method based on the correlation of the measurement indicators fully considers the volatility of the data,so that the detection result has higher accuracy.3.Based on the above research results,the data cleaning module in the power grid harmonic monitoring and analysis system is designed and implemented.