Detection Algorithm And Application Of Abnormal Electricity Behavior Based On Big Data

In recent years,with the rapid construction and development of smart grid,abnormal electricity usage is increasingly rampant,which seriously affects the operation of the grid system and the benefits of power supply enterprises.To research a new technique for detecting abnormal electricity behavior,not only provides the important decision support for electric power company,but also plays an important role in protecting residential electricity safety and maintain the electricity market.At the same time,it is also an inevitable trend of future development to analyze anomalous data,which has accumulated by intelligent terminal,in big data platform.The traditional single anomaly detection method is inefficient to process the large amount of electricity data with randomness and imbalance.Aiming at this problem,this thesis analyzes the user on the electrical behavior of different characteristics based on the electricity data,Combining the classification prediction algorithm and imbalanced data processing technology,we propose an algorithm dealing with imbalanced data classification based on random forest.Meanwhile,to cope with the exponential growth problem of the massive data,we realize a detection algorithm under big data platform,which greatly shortens the running time of the prediction algorithm.The main achievements of this thesis are as follows:1.Constructing the classification model of abnormal electricity usage based on user classifications.Firstly,we analyze the data set,preprocess the electricity data,and extract the behavior characteristics feature,establish the classification forecast model.Then,we compare the single classifier and the integrated learning method of abnormal user in this classify model.The experimental results show that the classification performance of random forest algorithm in this prediction model is superior than the traditional classification algorithm such as Decision Tree,Naive Bayes and KNN.2.Putting forward the application of imbalanced data classification algorithm based on random forest in the detection of abnormal electrical activities.Firstly,this thesis analyzes several factors that affect the performance of classifier,according to the characteristics of imbalanced distribution of power data sets,we propose an abnormal electric behavior detection algorithm based on imbalanced classification.Experiments are conducted to balance the electricity data,and to extract features and training model again.In the end,we contrast the classification results under several different imbalanced classification methods,which based on data processing and integration algorithm.The results show that,the random forest algorithm based on integrated imbalance has the best classification performance on this data set,3.Realizing th e abnormal electric behavior detection algorithm and application on Spark platform.We construct a classification prediction model based on big data platform.By using distributed computing framework,we design and implement random forest algorithm based on imbalanced classification.We contrast the average running time of several algorithms based on the stand-alone version and the Spark platform.The experimental results show that,without affecting the performance of algorithm classification,the average running speed of detection algorithm,which is implemented on the Spark platform,is higher than that of the standalone version about 20 times.The processing of massive electricity data based on the big data platform has promoted the running speed of the abnormal detection algorithm and shortened the detection time.