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

Nontechnical Loss Detection With Duffing–Holmes Self-Synchronization Dynamic Errors and 1D CNN-Based Multilayer Classifier in a Smart Grid

“…Such an ML method lacked the feature extraction function in data preprocessing. As shown in Table 5, DL-based methods such as wide and deep CNN, CNN-GRU, and CNN-LSTM [22,27,28,50,51], based on different datasets (for example, Smart Grid Corporation of China [21] or State Grid Corporation of China [22]), had also created a classifier to automate the feature extraction and classification processes for metering data classification, which could achieve greater than 85% classification accuracy and had higher indexes like Precision (%), Recall (%), and F1 score to evaluate the classifier's performance (as shown in Table 4). The DL-based approaches had a more complicated scheme, with many convolutionalpooling layers and a fully linked network to set up the various multilayer classifiers, which could automatically extract feature patterns without operator intervention and improve the classification accuracy.…”

“…The classic approaches mentioned above do not employ the quantizer to extract features and determine the appropriate feature parameters for boosting classification accuracy [20,21,29]. based methods, in contrast to the ML, the DP-based methods, The DP-based methods, in contrast to the ML-based methods, such as CNN and a long-shortterm memory (LSTM) model (CNN-LSTM), GoogLeNet and gated recurrent unit (GRU) model, wide and deep CNN, and CNN-GRU model [20,21,27,28], use the multi convolutionalpooling layers and a classification layer (fully connected layer) to learn the power consumption data in order to create a classifier that can detect electricity fraud, including autonomous end-to-end feature extraction, noise removal, and classification tasks. Their models can extract associated feature patterns from customers' consumption profiles using multi convolutional-pooling methods, which eliminates the need for more human participation and improves the classification accuracy rate.…”

A Novel Method CNN-LSTM Ensembler Based on Black Widow and Blue Monkey Optimizer for Electricity Theft Detection

“…This method has a low identification rate but a high rate of False Positives (FPR). [9] Missing Values Data…”

“…More analysis is necessary to address the issues of Electricity Theft Detection (ETD) adequately and overcome the constraints of inadequate theft detection owing to unbalanced data and the limited capacity of Machine Learning (ML) algorithms. We discovered that just a few publications in the current literature had addressed the impact of unbalanced data in their system models [9]. The authors in the literature solved the class imbalance issue by using Adaysn; however, this produces overfitting and repeats the samples of the closest neighbor, which will not represent theft cases of real-world [10].…”

An Energy-Fraud Detection-System Capable of Distinguishing Frauds from Other Energy Flow Anomalies in an Urban Environment

“…This requires a preliminary feature generation module to enable distinguishing between the two clusters. A work on cascading of Convolutional Neural Networks (CNN) and Gate Recurrent Unit (GRU) was performed by U. Ali et al [5]. The cascaded CNN-GRU algorithm reaches an average of 87% detection without the Manta Ray Foraging Optimization (MRFO) back-propagation algorithm, and up to 91% with MRFO.…”