FIGURE 3 - uploaded by Zeeshan Aslam
Content may be subject to copyright.
Source publication
Electricity theft is one of the main causes of non-technical losses and its detection is important for power distribution companies to avoid revenue loss. The advancement of traditional grids to smart grids allows a two-way flow of information and energy that enables real-time energy management, billing and load surveillance. This infrastructure en...
Context in source publication
Context 1
... and 3. The values of these plots depict the loss and accuracy of RDAE verses epochs during training. Fig. 2 shows the VOLUME 8, 2020 convergence of loss during the training and validation phases of RDAE. It demonstrates that RDAE consistently learns the power consumption patterns throughout feature extraction and uniformly minimizes the loss. In Fig. 3, we demonstrate the analysis of the proposed RDAE in terms of accuracy. It efficiently performs feature extraction and also derives features' associations. Afterwards, the weights are assigned to the extracted features by the AG and served as the unlabeled feature representations to AG-TripleGAN. On the other hand, the limited amount ...
Similar publications
This paper presents a hybrid model, named as hybrid deep neural network, which combines convolutional neural network, particle swarm optimization, and gated recurrent unit, termed as convolutional neural network-particle swarm optimization-gated recurrent unit model. The major aims of the model are to perform accurate electricity theft detection an...
Citations
... A transductive SVM (TSVM) method [79] has been utilized as a semi-supervised learning for ETD, but it may encounter challenges in scaling when confronted with large volumes of data. In deep learning-based semi-supervised learning for ETD, two primary approaches can be employed: (1) augmenting data by assigning pseudo-labels [80,81] and (2) integrating supervised and unsupervised learning [82,83]. In the first approach, a model can assign pseudo-labels to unlabeled data, reducing overfi ing and improving model generalization [80,81]. ...
... In deep learning-based semi-supervised learning for ETD, two primary approaches can be employed: (1) augmenting data by assigning pseudo-labels [80,81] and (2) integrating supervised and unsupervised learning [82,83]. In the first approach, a model can assign pseudo-labels to unlabeled data, reducing overfi ing and improving model generalization [80,81]. In the second approach, networks trained on unsupervised and supervised learning tasks are used to reconstruct load profiles and differentiate between classes [82,83]. ...
The transition to smart grids has served to transform traditional power systems into data-driven power systems. The purpose of this transition is to enable effective energy management and system reliability through an analysis that is centered on energy information. However, energy theft caused by vulnerabilities in the data collected from smart meters is emerging as a primary threat to the stability and profitability of power systems. Therefore, various methodologies have been proposed for energy theft detection (ETD), but many of them are challenging to use effectively due to the limitations of energy theft datasets. This paper provides a comprehensive review of ETD methods, highlighting the limitations of current datasets and technical approaches to improve training datasets and the ETD in smart grids. Furthermore, future research directions and open issues from the perspective of generative AI-based ETD are discussed, and the potential of generative AI in addressing dataset limitations and enhancing ETD robustness is emphasized.
... Presently, both the developing and developed nations are the victims of electricity fraud. More specifically, the USA loses a profit of $6 billion annually in the account of electricity theft [4], both India and Brazil face a loss of about $4.5 billion per year [5], the UK loses $173 million yearly from its profit [6], Canada is suffering from an annual loss of $100 million, and Pakistan experiences a loss of 17.03% of its total energy because of electricity fraud [7,8]. Moreover, the other growing nations' utilities face almost a loss of 50% of their profit due to electricity theft [9]. ...
One of the crucial issues for power grids in strengthening the urbanization around the world is imbalance between supply and demand, which leads the users to consume electricity in an anomalous manner without paying for it. Electricity theft plays a pivotal role in cutting down on the electricity bills. The existing data-oriented approaches for electricity theft detection (ETD) in the smart cities have limited ability to handle noisy high-dimensional data and features’ associations. These limitations raise the misclassification rate, which makes some of the approaches unacceptable for electric utilities. A new twofold end-to-end methodology is proposed for ETD. In the first fold, it groups the similar electricity consumption (EC) cases through grey wolf optimization (GWO)-based clustering mechanism; clustering by fast search and find of density peaks (CFSFDP), we named it GC. In the second fold, a new relational stacked denoising autoencoder (RSDAE)-based semi-supervised generative adversarial network (GAN), termed as RGAN, is used for ETD. The combined methodology is named as GC-RGAN. In the methodology, RSDAE acts as both feature extraction technique and generator sub-model of the proposed RGAN. The proposed methodology utilizes the advantages of clustering, adversarial learning and semi-supervised EC data. Besides, to validate the effectiveness of the proposed solution, extensive simulations are performed using smart meter data. Simulation results validate the excellent ETD performance of the proposed GC-RGAN against existing ETD schemes, such as random forest and semi-supervised support vector machine. In comparison, GC-RGAN covers the ETD score of 98% that shows its suitability for real-world scenarios. The proposed solution has extraordinary performance for ETD as compared to traditional solutions, which shows its superiority and usefulness for real-world applications.
... Subsequently, the decoder takes this internal representation generated by the encoder as the source information and endeavors to reconstruct the original input. Examples of autoencoder applications can be found in [91][92][93][94][95][96][97][98]. ...
Nontechnical losses of electrical energy (NTLEE) have been a persistent issue in both the Russian and global electric power industries since the end of the 20th century. Every year, these losses result in tens of billions of dollars in damages. Promptly identifying unscrupulous consumers can prevent the onset of NTLEE sources, substantially reduce the amount of NTLEE and economic damages to network grids, and generally improve the economic climate. The contemporary advancements in machine learning and artificial intelligence facilitate the identification of NTLEE sources through anomaly detection in energy consumption data. This article aims to analyze the current efficacy of computational methods in locating, detecting, and identifying nontechnical losses and their origins, highlighting the application of neural network technologies. Our research indicates that nearly half of the recent studies on identifying NTLEE sources (41%) employ neural networks. The most utilized tools are convolutional networks and autoencoders, the latter being recognized for their high-speed performance. This paper discusses the main metrics and criteria for assessing the effectiveness of NTLEE identification utilized in training and testing phases. Additionally, it explores the sources of initial data, their composition, and their impact on the outcomes of various algorithms.
... Alternatively, unsupervised methods used for ETD are entropy-based detection [17], K-means clustering-based model [18], self-organizing map (SOM) [4], fuzzy logic or clustering [19], [20], LSTM-Gaussian mixture model [21], Markov-chain model [22], autoencoders [23], etc. Moreover, there are also semi-supervised methods that use both labeled and unlabeled data to detect inspected and un-inspected theft cases [24][25][26][27]. Most of these models are less accurate in terms of ETD with high computational time and dependence on the domain knowledge to perform feature selection and extraction. ...
Electricity theft is the largest type of non-technical losses faced by power utilities around the globe. It not only raises revenue losses to the utilities but also leads to lethal fires and electric shocks at distribution side. In the past, field operation groups were sent by the utilities to conduct inspections of suspicions electric equipments stated by the public. Advanced metering infrastructure based recent development in the smart grids makes it easy to detect electricity thefts. However, the conventional supervised learning techniques have low theft detection performance mainly due to imbalance datasets available for training. Therefore, in this paper, we develop a novel theft detection model with twofold contribution. A unique hybrid sampling technique named as hybrid oversampling and undersampling using both classes (HOUBC) is proposed to balance the dataset. HOUBC first performs undersampling and then oversampling using both the majority (normal) and minority (theft) classes. A new deep learning method, fractal network is applied with light gradient boosting method to extract and learn important characteristics from electricity consumption profiles for identifying electricity thieves. The proposed model relies on smart meter's data for theft detection and hence, a rapid and widespread adaption of this model is feasible, which shows its main advantage. The performance of the model is evaluated with real-world smart meter's data, i.e., state grid corporation of China. Comprehensive simulation results describe the effectiveness of the proposed model against conventional schemes in terms of electricity theft detection.
... Specifically, the inclusion of unlabeled data allows the SSL algorithm to infer close-to-optimal decision boundaries, thereby improving classification accuracy. Previous studies have applied SSL techniques for fault diagnosis and event detection, 103,104 detection of cyber-attacks such as false data injection attacks, 105 electricity theft detection, 106 and nonintrusive load monitoring. 107,108 However, the full potential of SSL has yet to be explored in various distribution system applications. ...
The application of machine learning (ML) to power and energy systems (PES) is being researched at an astounding rate, resulting in a significant number of recent additions to the literature. As the infrastructure of electric power systems evolves, so does interest in deploying ML techniques to PES. However, despite growing interest, the limited number of reported real-world applications suggests that the gap between research and practice is yet to be fully bridged. To help highlight areas where this gap could be narrowed, this article discusses the challenges and opportunities in developing and adapting ML techniques for modern electric power systems, with a particular focus on power distribution systems. These systems play a crucial role in transforming the electric power sector and accommodating emerging distributed technologies to mitigate the impacts of climate change and accelerate the transition to a sustainable energy future. The objective of this article is not to provide an exhaustive overview of the state-of-the-art in the literature, but rather to make the topic accessible to readers with an engineering or computer science background and an interest in the field of ML for PES, thereby encouraging cross-disciplinary research in this rapidly developing field. To this end, the article discusses the ways in which ML can contribute to addressing the evolving operational challenges facing power distribution systems and identifies relevant application areas that exemplify the potential for ML to make near-term contributions. At the same time, key considerations for the practical implementation of ML in power distribution systems are discussed, along with suggestions for several potential future directions.
... Different from two-stage methods, one-stage methods identify electricity theft based on the data directly. Many one-stage methods employ a single classifier, such as support vector description [21], LightGBM [22,23], XGBoost [23,24], long short-term memory (LSTM) [25], deep vector embedding [26], Bayesian risk framework [27], linear regression [28], the black hole algorithm [29], random forest [30], feed forward neural networks [31], TripleGAN [32], etc. Note that [30][31][32] use a stacked autoencoder, deep autoencoder, and relational autoencoder for features extraction, and then use a random forest, feed forward neural networks, and TripleGAN for classification. ...
... Many one-stage methods employ a single classifier, such as support vector description [21], LightGBM [22,23], XGBoost [23,24], long short-term memory (LSTM) [25], deep vector embedding [26], Bayesian risk framework [27], linear regression [28], the black hole algorithm [29], random forest [30], feed forward neural networks [31], TripleGAN [32], etc. Note that [30][31][32] use a stacked autoencoder, deep autoencoder, and relational autoencoder for features extraction, and then use a random forest, feed forward neural networks, and TripleGAN for classification. However, because the main function of the autocoder is dimensionality and noise reduction, which can be regarded as a data preprocessing process, [30][31][32] are still regarded as one-stage algorithms in this paper. ...
... Note that [30][31][32] use a stacked autoencoder, deep autoencoder, and relational autoencoder for features extraction, and then use a random forest, feed forward neural networks, and TripleGAN for classification. However, because the main function of the autocoder is dimensionality and noise reduction, which can be regarded as a data preprocessing process, [30][31][32] are still regarded as one-stage algorithms in this paper. Meanwhile, some of the one-stage methods fuse a multi-classifier to further improve the performance of the detection methods. ...
This paper proposes a novel convolution–non-convolution parallel deep network (CNCP)-based method for electricity theft detection. First, the load time series of normal residents and electricity thieves were analyzed and it was found that, compared with the load time series of electricity thieves, the normal residents’ load time series present more obvious periodicity in different time scales, e.g., weeks and years; second, the load times series were converted into 2D images according to the periodicity, and then electricity theft detection was considered as an image classification issue; third, a novel CNCP-based method was proposed in which two heterogeneous deep neural networks were used to capture the features of the load time series in different time scales, and the outputs were fused to obtain the detection result. Extensive experiments show that, compared with some state-of-the-art methods, the proposed method can greatly improve the performance of electricity theft detection.
... Therefore, it is necessary to handle the missing values and outliers through preprocessing techniques. In this regard, we apply the concept of linear interpolation to recover the missing values in the dataset [58], which is numerically illustrated as: ...
The reduction of electricity theft in power sectors is an important concern of utilities as it represents an essential part of their total potential benefits. The existing electricity theft detection (ETD) approaches have unsatisfactory performance due to high dimensional and imbalanced data. Moreover, the existing approaches also have ineffective results due to the limited availability of supervised data and auxiliary information. Therefore, we introduce a new mechanism that is based on two scenarios. In the first scenario, a new supervised learning solution is presented, which is a combination of UNet and generative adversarial network (GAN), named as UNet-GAN. The GAN's structure is mainly comprised of two neural networks: generator and discriminator. Due to an excellent performance of UNet, we utilize it in both generator and discriminator parts. These two neural networks contest with each other in a game-theoretic manner to significantly boost the ETD performance. In the second scenario, a novel dynamic learning based semi-supervised solution is proposed that consists of probabilistic guider (PG) and Ladder network. The solution is termed as PG-Ladder network. PG dynamically guides the proposed network to further improve its performance in terms of ETD. Furthermore, the performance of the proposed solution is evaluated over the suitable classification indicators using real electricity consumption (EC) records. The results indicate more efficient performance of the proposed solution than the state-of-the-art approaches regarding the ETD.
... One of the main objectives of ETD is to enhance the TPR or Detection Rate (DR) while simultaneously reducing the FPR. Thus, the ROC-AUC is a useful metric for identifying NTLs in binary classification problems [33]. It demonstrates the relationship of TPR with FPR at different threshold values. ...
The current study uses a data-driven method for Nontechnical Loss (NTL) detection using smart meter data. Data augmentation is performed using six distinct theft attacks on benign users’ samples to balance the data from honest and theft samples. The theft attacks help to generate synthetic patterns that mimic real-world electricity theft patterns. Moreover, we propose a hybrid model including the Multi-Layer Perceptron and Gated Recurrent Unit (MLP-GRU) networks for detecting electricity theft. In the model, the MLP network examines the auxiliary data to analyze nonmalicious factors in daily consumption data, whereas the GRU network uses smart meter data acquired from the Pakistan Residential Electricity Consumption (PRECON) dataset as the input. Additionally, a random search algorithm is used for tuning the hyperparameters of the proposed deep learning model. In the simulations, the proposed model is compared with the MLP-Long Term Short Memory (LSTM) scheme and other traditional schemes. The results show that the proposed model has scores of 0.93 and 0.96 for the area under the precision–recall curve and the area under the receiver operating characteristic curve, respectively. The precision–recall curve and the area under the receiver operating characteristic curve scores for the MLP-LSTM are 0.93 and 0.89, respectively.
... The semi-supervised learning method uses a small amount of label data obtained to train the initial learner, test and classify the unknown category data, and add the samples with high confidence coefficient in the classification results to the training set to train the model again, and repeat this process until all samples are the most excellent classification. Reference [19] uses a correlation denoising autoencoder to achieve feature extraction and feature association of electricity data. In [20], the authors propose a semi-supervised learning-based SSAE generation model and design an adversarial module to enhance the model's anti-noise ability. ...
Nowadays, electricity theft has been a major problem worldwide. Although many single-classification algorithms or an ensemble of single learners (i.e., homogeneous ensemble learning) have proven able to automatically identify suspicious customers in recent years, after the accuracy of these methods reaches a certain level, it still cannot be improved even if it continues to be optimized. To break through this bottleneck, a heterogeneous ensemble learning method with stacking integrated structure of different strong individual learners for detection of electricity theft is presented in this paper. Firstly, we use the grey relation analysis (GRA) method to select the heterogeneous strong classifier combination of LG + LSTM + KNN as the base model layer of stacking structure based on the principle of the highest comprehensive evaluation index value. Secondly, the support vector machine (SVM) model with relatively good results of the stacking overall structure experiment is selected as the model of the meta-model layer. In this way, a heterogeneous integrated learning model for electricity theft detection of the stacking structure is constructed. Finally, the experiments of this model are conducted on electricity consumption data from State Grid Corporation of China, and the results show that the detection performance of the proposed method is better than that of the existing state-of-the-art detection method (where the area under receiver operating characteristic curve (AUC) value is 0.98675).
... The work by Cardenas et al. focused on theft detection as a zero-sum game and used Nash's equilibrium condition to detect some parameters that deferred between agent #1 "representing customer" winning or agent #2 winning (representing the company). There was a preliminary idea to represent each side by an agent and use a "generative adversarial network" (GAN) architecture [10], since GAN is a zero-sum game implementation and using GAN for theft detection, not for data augmentation, as presented in [11]. Two major problems with this idea are as follows: (1) there are no representative data available to the customer. ...
This paper describes an electricity technical/nontechnical loss detection method capable of loss type identification, classification, and location. Several technologies are implemented to obtain that goal: (i) an architecture of three generative cooperative AI modules and two additional non-cooperative AI modules for data knowledge sharing is proposed, (ii) new expert consumption-based knowledge of feature collaboration of the entire consumption data are embedded as features in an AI classification algorithm, and (iii) an anomaly pooling mechanism that enables one-to-one mapping of signatures to loss types is proposed. A major objective of the paper is an explanation of how an exact loss type to signature mapping is obtained simply and rapidly, (iv) the role of the reactive energy load profile for enhancing signatures for loss types is exemplified, (v) a mathematical demonstration of the quantitative relationship between the features space to algorithm performance is obtained generically for any algorithm, and (vi) a theory of “generative cooperative modules” for technical/nontechnical loss detection is located and mapped to the presented system. The system is shown to enable high-accuracy technical/nontechnical loss detection, especially differentiated from other grid anomalies that certainly exist in field conditions and are not tagged in the universal datasets. The “pooling” architecture algorithm identifies all other loss types, and a robotic process automation module obtains loss type localization. The system feeds from the entire smart metering data, not only the energy load profile. Other solutions, such as a stand-alone algorithm, have difficulty in obtaining low false positive in field conditions. The work is tested experimentally to demonstrate the matching of experiment and theory.
