Research On Abnormal Electricity Consumption Detection Method For Smart Grid Users Based On Deep Learning

For a long time,electricity theft has been a common phenomenon in power grids,which not only brings huge economic losses to power grid companies,but also poses great potential risks to users’ equipment and personal safety,so it has become an urgent problem to realize the detection of electricity users’ electricity abnormality and accurate audit of electricity theft users.With the continuous development of Advanced Metering Infrastructure of Smart Grid and the popularization of Smart Meters,power grid companies can collect more comprehensive and accurate data from the grid,so that they can use Artificial Intelligence technology to enhance the level of intelligent analysis and decision making of the grid.In this paper,we have conducted a research and system implementation of abnormal electricity users detection method based on Deep Learning for smart grid,and the main work is as follows.1.For the problem of data imbalance between normal and abnormal users in the abnormal electricity consumption detection,this paper proposes a Time GAN-based data enhancement technique to balance the dataset.In the Deep Learning model,the learning effect of features for abnormal users is poor if there is a huge difference between normal and abnormal user data,and the existing data generation methods cannot learn the distribution and time series features of the original data well,while the Time GAN-based data augmentation technique proposed in this paper to generate user electricity theft data can not only capture the time series within each time point of distribution,it also captures the potentially complex dynamic features of electricity consumption data across time.Experiments demonstrate that using Time GAN-based generated data input to the abnormal electricity consumption detection model can effectively improve the precision rate and reduce the false detection rate,and in order to demonstrate the superiority of Time GAN-generated data,this paper uses PCA and t-SNE techniques to compare other data balancing methods and qualitatively assess the degree of fit of the generated data.2.For the problem of low precision rate and high false detection rate of existing detection methods,this paper proposes a hybrid multi-time-scale neural network based abnormal electricity consumption detection model.Most of the existing electricity theft detection models use shallow machine learning techniques,which do not make full use of the time-series features of electricity consumption sequences and ignore the differences in time series correlation between normal and abnormal users.In this paper,based on the statistical analysis and correlation analysis of electricity consumption data of normal and abnormal users,we especially construct an abnormal electricity consumption detection model based on hybrid multi-time-scale neural network,which extracts the global features and deep periodic features of electricity consumption data using Bi-LSTM,Res Net,and Alex Net at three time scales of daily,weekly,and monthly,respectively,and uses Ada Boost classifier to achieve user classification with high precision and low false detection rate.For the above proposed method,real electricity consumption data provided by the State Grid Corporation of China are used for training and testing,and experiments are conducted to compare various balancing data methods,compare experiments with other models and model ablation experiments,and then verify the effectiveness of the method in this paper.3.For the problem of how to precisely identify specific users of electricity theft,this paper designs and implements an abnormal electricity consumption detection system.In order to bridge the gap between the lack of knowledge and the technological gap of the power grid staff,this paper builds an abnormal electricity consumption detection system based on a hybrid multi-time-scale neural network-based abnormal electricity consumption detection model,which can be applied by the power grid inspection staff to regularly check the users in each area,screen the suspected users,and then target them precisely and quickly to combat electricity theft.