Electricity Theft Detection using CNN-GRU and Manta Ray Foraging Optimization Algorithm

Besides the non-technical losses of power companies, theft of electricity is the most serious and dangerous one. The fraudulent power consumption degrades the quality of supply and increases the energy generation that impacts the whole grid system, which causes the legitimate users to pay a huge amount of electricity bills. Through data analysis methods, Smart Grid (SG) adaptation can significantly reduce this loss. SG infrastructure produces large amounts of data, including electricity consumer consumption. Machine learning and deep learning methods are using this historical record of user's data and can identify who steals electricity. Theft detection system is proposed in this paper, which consists of Convolutional Neural Network (CNN) and Gated Recurrent Unit (GRU) architecture. CNN is a widely used technology that can perform feature extraction automatically. Moreover, CNN also performs classification process with the power consumption of time series data. To perform classification in smart grid, we established a CNN-based GRU (CNN-GRU) model. Also, the hyper parameters of CNN-GRU are set tuned with a swarm based optimization algorithm Manta Ray Foraging Optimization (MRFO). Further, a pre-processing algorithm is implemented to calculate the missing values in the dataset, which is based on the local value associated with the missing data point. Also in this dataset, relatively few users have power theft and the model may not be able to effectively identify the steal users. Such imbalances were solved by comprehensive data generation using Synthetic Minority Over-sampling Technique (SMOTE). In the end, the obtained results show that our proposed technique can better classify most categories (normal users) and most minorities (electric theft users). The accuracy of CNN-GRU-MRFO is 91.1%, which is 6% higher than actual CNN-GRU.