Research On Smart Grid Big Data Outlier Detection And Analysis Of Electricity Behavior Based On Density Peaks Clustering Algorithm

With the construction and development of smart grid,all aspects of the smart grid are producing huge volume,complex structure and complex correlation data,which are the main source of smart grid big data.The value of the data is generated in data analysis.It can play an important role in the aera of grid operation,power grid assets and electricity users through the analysis and mining of the smart grid big data.Feature extraction and clustering analysis are the basis of smart grid big data analysis,and they are the key factors influencing the analysis results.In addition to the knowledge of business domain,it is necessary to understand the background knowledge of statistical and machine learning.For the feature extraction algorithm,this thsis compares the discrete wavelet transform and the gaussian mixture model algorithm,and gives the reason why the discrete wavelet transform is used in the analysis of electricity behavior.For the clustering algorithm,this thsis compares K-Means,DBSCAN and find of density peaks clustering algorithm,and gives the reasons for choosing the fast density peak clustering algorithm to improve and used to outlier detection and user behavior analysis.Due to data sources,statistical caliber,personnel entry,abnormal behavior and the lack of data quality control system,will lead to outlier data.Outlier data contains information about the appearance of the system anomaly,so it has a huge value to research on outlier data.Meanwhile,the existence of outlier data will affect the accuracy of feature extraction and clustering.Therefore,this thsis proposed a fast search and find of density peaks outlier detection algorithm based on KNN.In order to overcome the shortcomings in fast search and find density peaks clustering algorithm,i)the local characteristics of the datasets are not considerd and ii)the local density is sensitive to the cut-off distance,the idea of KNN algorithm is used to redefine the local density and distance to achieve more accurate outlier detection.Moreover,the rules to determine the outliers is designed.Therefore,the proposed algorithm could achieve more accurate outlier detection by improving the drawbacks that the original algorithm does not take into account the local characteristics of the data and relies on the cut off distance.Simulation results based on the daily load data of the distribution transformers in a certain province demonstrate the effectiveness of the algorithm.Electricity behavior analysis is an important part of smart grid big data,which is the basis of load forecasting,demand response,grid planning and operation,rate setting,energy efficiency and so on.In addition to using the idea of KNN to redefine the local density and distance,outward statistical test method is adopted to identify clustering centers automatically to overcome the disadvantage of fast search and find of density peaks clustering algorithm needs identifying clustering centers by two empirically preassigned minimum thresholds.Discrete wavelet transform(DWT)is used to extract the multi-time scale characteristics of load profile,and then the load data of different time scales are clustered and the typical load curve is reconstructed to realize the analysis of electricity behavior.The analysis method has achieved good results on the actual data set of individual users and users in different industries.