Research On Intelligent Electric Meters Fault Prediction Technology Based On Big Data Analysis

Compare with the traditional meter with the basic electricity measure function,it also has a two-way multi-rate measurement,the user control,two-way data communications capabilities,anti-theft and other intelligent features,Majority of maintenance personnel do not have the ability to repair all types of faults,therefore,when the smart meter fault,if we can accurately predict the type of the meter and guide to dispatch the maintenance personnel,which will effectively reduce the state grid operation and maintenance costs and resource costs.First of all,we analyzed the relationship between the various features of the smart meter and the type of the fault.According to the fault data collected in recent years,the histogram and proportional summation map are used to analyze the distribution of fault data and analyzed the possible causes of the failure of the smart meter,and provided the basis for the data preprocessing method.Secondly,different preprocessing methods are designed for the category and continuous features of the smart meter data.Based on the many categories of categorization features,this paper proposes a clustering method based on hierarchical clustering to combine categories feature with similar fault distribution,to reduce the data dimension,and then do binary processing.In order to solve the problem that there are many missing values in continuous data,this paper adopts the data complement method based on feature distribution function,then the matrix is compressed by using the sparseness of the matrix after processing.Thirdly,the research on smart grid fault prediction technology is carried out.We have been done lots of efforts on the technology of predict smart meter fault type.First of all,a deep neural network that can well fit the relationship between input and output is designed,and the parameters such as the optimization method are debugged.Secondly,due to the imbalance of fault type,a cost-sensitive XGBoost model is designed,and shows good effect on the test data.Finally,in order to solve the relatively large redundancy in the fault data,this paper proposes a XGBoost fault predict method based on weighted column subsampling.After analysis and study,adding redundant features in the training model will not only increase the training time but also affect the accuracy of the model.Therefore,we proposed when doing column subsampling during training each tree,the probability of each feature to be drawn is proportional to its feature importance,which effectively improves the efficiency of the ensemble algorithm in training data with large redundancy.Futhermore,our proposed method is general,which can be easily extended to other methods which need column subsampling,when the feature of data is high dimensional and redundant.The validity of the method is verified on meter faulty data.