Electricity Consumption Data Quality Improvement Method For New Power System Data-driven Advanced Applications

The new power system characterized by the efficient use of clean energy,through short-term load forecasting and demand response,realizes flexible interaction of power distribution and orderly power consumption,and promotes users to save energy and reduce emissions.User power consumption data is the basis for various advanced applications such as short-term load forecasting,station line loss management,and antielectricity theft.However,the actual power grid electricity information collection system has a complex working environment,and data quality problems such as missing electricity data,noise and abnormalities are inevitable,which seriously affects the effects of various advanced applications.Therefore,it is of great significance to study the methods to improve the quality of electricity consumption data for marketing measurement.In response to the above problems,combined with actual marketing measurement scenarios,the characteristics of marketing measurement electricity consumption data is analyzed,and a matrix clustering transformation is proposed to reduce the rank of the electricity consumption data matrix,and summarize the characteristics and causes of various data quality problems such as missing data,noise,and anomalies.The data quality improvement method based on matrix norm optimization proposed in this thesis first restores the missing data by minimizing the matrix kernel norm model,and then realizes matrix clustering transformation through density-based mean shift clustering.On this basis,consider Gaussian white noise,spike noise and abnormal data,improve the classic low-rank sparse decomposition model,establish a multi-norm optimized low-rank matrix repair model,and repair missing data,noise,and abnormal data at the same time to improve data quality.Considering the efficiency of data restoration,the solution algorithm of this model is designed based on the alternating direction multiplier method.The problem to be solved is decomposed into the missing data recovery sub-problem,the Gaussian white noise extraction sub-problem,the spike noise extraction sub-problem and the abnormal data extraction sub-problem.And the iterative solution steps of each sub-problem are obtained respectively.At the same time,an adaptive iterative step is designed.The length of step selection strategy accelerates the model solving speed and improves the accuracy and efficiency of model solving.Finally,the experimental examples in this thesis are based on real user electricity data,and carry out missing data repair experiments,noisy abnormal data repair experiments,and missing and noisy data mixed scene repair experiments.Compared with other algorithms,it is proved that the method in this thesis can improve the accuracy of missing data recovery and improve data quality.And the short-term load forecasting experiment based on artificial intelligence LSTM method proves that the data quality restoration method proposed in this thesis can effectively improve the accuracy and effect of short-term load forecasting,and has good practical significance for the new advanced applications of power system based on data-driven.