Skip to main content An official website of the United States government Here's how you know Here's how you know Official websites use .gov A .gov website belongs to an official government organization in the United States. Secure .gov websites use HTTPS A lock ( Lock Locked padlock icon ) or https:// means you've safely connected to the .gov website. Share sensitive information only on official, secure websites. Search Log in Dashboard Publications Account settings Log out Search… Search NCBI Primary site navigation Search Logged in as: Dashboard Publications Account settings Log in Search PMC Full-Text Archive Search in PMC Journal List User Guide PERMALINK Copy As a library, NLM provides access to scientific literature. Inclusion in an NLM database does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more: PMC Disclaimer | PMC Copyright Notice Sci Rep . 2026 Mar 17;16:10515. doi: 10.1038/s41598-026-38824-w Search in PMC Search in PubMed View in NLM Catalog Add to search Reinforcing smart grid resilience through blockchain-supported deep learning models for theft detection Fadia Bibi Fadia Bibi 1 University Institute of Information Technology, PMAS-Arid Agriculture University, Rawalpindi, Pakistan Find articles by Fadia Bibi 1 , Saif Ur Rehman Saif Ur Rehman 1 University Institute of Information Technology, PMAS-Arid Agriculture University, Rawalpindi, Pakistan Find articles by Saif Ur Rehman 1, ✉ , Sarfraz Bibi Sarfraz Bibi 1 University Institute of Information Technology, PMAS-Arid Agriculture University, Rawalpindi, Pakistan Find articles by Sarfraz Bibi 1 , Kaman Aziz Kaman Aziz 2 Department of Physical and Numerical Sciences, Qurtuba University of Science & Information Technology, D.I. Khan, Pakistan Find articles by Kaman Aziz 2 , Ahmad Alshammari Ahmad Alshammari 3 Department of Computer Sciences, Faculty of Computing and Information Technology, Northern Border University, 91911 Rafha, Saudi Arabia Find articles by Ahmad Alshammari 3 , Vincent Karovič Vincent Karovič 4 Department of Information Management and Business Systems, Faculty of Management, Comenius University, Bratislava, Slovakia Find articles by Vincent Karovič 4, ✉ Author information Article notes Copyright and License information 1 University Institute of Information Technology, PMAS-Arid Agriculture University, Rawalpindi, Pakistan 2 Department of Physical and Numerical Sciences, Qurtuba University of Science & Information Technology, D.I. Khan, Pakistan 3 Department of Computer Sciences, Faculty of Computing and Information Technology, Northern Border University, 91911 Rafha, Saudi Arabia 4 Department of Information Management and Business Systems, Faculty of Management, Comenius University, Bratislava, Slovakia ✉ Corresponding author. Received 2025 Sep 27; Accepted 2026 Jan 31; Collection date 2026. © The Author(s) 2026 Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/ . PMC Copyright notice PMCID: PMC13035815 PMID: 41844675 Abstract With the increasing complexity of smart grid data, detecting fraudulent activities such as electricity theft has become increasingly challenging. Smart grids facilitate real-time monitoring, providing a valuable platform for identifying anomalous consumption patterns. To address this issue, we propose a deep learning-based framework that integrates an LSTM-Autoencoder model for electricity theft detection. The model effectively captures long-term temporal dependencies and identifies persistent anomalies, enhancing the standard LSTM architecture’s capability to model sequential data. For robust security and transparency, the framework incorporates blockchain technology, establishing a decentralized logging mechanism that prevents data tampering and ensures a trustworthy audit trail. This integration enables secure, verifiable, and transparent recording of detected anomalies and operational events. Furthermore, the proposed approach is implemented in Python using deep learning frameworks such as TensorFlow and Keras, with optional PyTorch support. Extensive experiments demonstrate that the combined LSTM-Autoencoder and blockchain framework achieves 95% accuracy, outperforming traditional and hybrid detection methods. The solution is scalable, privacy-preserving, and provides a resilient, intelligent, and transparent ecosystem for smart grid operations, offering a significant advancement in electricity theft detection and operational reliability. Supplementary Information The online version contains supplementary material available at 10.1038/s41598-026-38824-w. Keywords: Electricity theft, Smart grid, LSTM-autoencoder, Blockchain integration, Data privacy and integrity, Temporal dependencies Subject terms: Energy science and technology, Engineering, Mathematics and computing Introduction The smart grid plays a significant role in modern energy systems by integrating multiple energy sources, including both conventional and renewable solutions 1 , 2 . The traditional electrical power grid which was unable to satisfy requirement of the 21 st century, as its basic framework remained largely unchanged despite increasing demand 3 . Population growth, along with raising electricity consumption, has placed immense pressure on the existing grid infrastructure, which was built using obsolete facilities and technologies 4 . As a result traditional electricity power supply system have given rise to numerous operational challenges, leading to the emergence of a new paradigm known as the smart grid 5 . The smart grid represents an advance electrical power distribution network that integrates energy flow with data exchange in a bi-directional manner, enabling a computerized and centralized approach to power management 6 , 7 . The primary objective of the smart grid is to ensure sustainability and reliability of electricity supply while simultaneously enabling active participation from producers and consumers both generate and consume electricity 8 . When compared to the classical power system, where electricity is delivered from centralized substation to various consumer types 9 , smart grid adhere to a more distributed and intelligent operational model 9 , 10 . Over the past few years, the economy of Canada, United Kingdom and the United States has been estimated to lose about U.S $100 million 6 , U.S $170 million and U.S $6 billion 8 respectively and in Pakistan, it is estimated between 20 and 30% of the electricity produced in technical inefficiencies and theft. This results in annual financial losses ranging from PKR 500 billion to 600 billion 11 . During the fiscal year 2022–23, distribution companies reported losing of over 19.17 billion units accounting for approximately 16.4% of the electricity purchased from generation companies due to transmission losses as well as electricity theft 12 , 13 . In addition to direct economic losses, electricity theft also disrupts energy distribution and generation processes 14 .Consequently, effective and comprehensive energy theft detection has remained a major concern in the smart grid systems 15 . Furthermore, traditional electricity theft detection methods used in conventional power grids are often time consuming and inefficient in the context of modern technological advancements 16 . In smart grid, energy consumption data can be collected automatically 17 , 18 , detecting electricity theft detection executing traditional methods machine learning models used Support vector machine (SVM), K-Nearest Neighbor(KNN), Logistic Regression(LR), Neural Network (CNN,RNN), Bayes, Decision Tree, Artificial Neural Network are commonly used 19 . Also, more advanced deep learning models have been experimented, which are included Deep Neural Networks (DNN), Convolution Neural Network(CNN), Long Short-Term Memory (LSTM) 20 , Recurrent Neural Network (RNN, including Simple Recurrent Unit and Bi-directional Recurrent Unit), Stacked Auto encoder , and Gated Recurrent Units (GRU) offer enhanced representations and performance. Furthermore, some ensemble approaches have also been proposed to improve detection accuracy, often combined with blockchain integration for enhance security 10 , 21 , 22 . Security in different domains such as in edge computing and Blockchain are of prime significance and it provides better solution for electricity theft detection by providing a secure, decentralized pattern to verify energy consumption data across grid 23 . Blockchain based framework adopted in smart grid make electricity system smarter and safer 24 – 26 . In real time monitoring, various techniques are adopted allowing utility providers to detect unusual consumption patterns and identify the theft consumers in smart grid 27 . Further, actual consumption data can be utilized by the smart contracts to compute the invoices 28 , 29 , thus minimizing the fraud and invoicing disagreements through the use of the Blockchain technology 30 . The Blockchain technique has decentralized methods that support peer-to-peer trading of energy within a distributed energy system environment 31 , 32 . It is required by data privacy and security within a smart grid using Blockchain technology 33 , 34 . The case to establish robust security systems is in line with the integration of blockchain technology into the Internet of Things (IoT) 35 . A typical Blockchain-based electricity theft detection framework has been presented in Fig. 1 . Fig. 1. Open in a new tab Blockchain-based electricity theft detection 52 . To address electricity theft, this paper proposes a blockchain-based anomaly detection framework for electricity consumption that supports distributed and collaborative anomaly detection. In this study, we propose a novel electricity theft detection approach that integrates blockchain technology, federated learning, and smart contracts to ensure privacy preservation and transparency. Smart meters locally store and analyze consumption data to detect anomalies using federated learning, while keeping raw usage data private and undisclosed. Only model updates are shared and recorded on an immutable blockchain ledger, ensuring tamper-proof logging and eliminating the risk of data manipulation. A blockchain-based incentive mechanism rewards consumers with honest usage patterns, while electricity theft is addressed through smart contracts that automatically trigger predefined actions, such as issuing alerts, imposing fines, or disconnecting services. As compared to existing centralized detection approaches, the proposed approach enhances privacy, transparency, automation of enforcement, and community participation. Moreover, it offers improved scalability, security, and trustworthiness for large-scale power grids. This anomaly detection mechanism not only enables timely and accurate identification of irregularities but also assists industry stakeholders in gaining a holistic view of the network and making informed, responsive decisions. Consequently, the major contributions of the presented study are as follow: Firstly, this study introduces an unsupervised sequence-to-sequence LSTM-Autoencoder algorithm that learns normal electricity consumption patterns and detects theft using reconstruction error. This approach effectively addresses severe data imbalance and captures long-term temporal dependencies in smart grid data. Secondly, a lightweight blockchain-aware architecture is proposed, where anomaly detection is performed off-chain, while only validated detection results (labels, reconstruction errors, and timestamps) are recorded on-chain. This design significantly reduces computational overhead and overcomes scalability limitations common in existing blockchain-centric approaches. Thirdly, this research develops custom smart contract logic that encodes anomaly validation, immutable event logging, and automated incentive/penalty enforcement. This enables tamper-proof auditing and trustworthy decision-making without exposing raw consumption data. The rest of this paper is organized as follows: Section “ Literature review ” presents a detailed review of recent work related to the scope of this study. Section “ Novel blockchain approach for electricity theft detection ” proposes a novel approach for electricity theft detection based on the identified research gaps, along with the associated algorithms. Section “ Experimental Results and Analysis ” provides a comprehensive analysis and simulation results. Finally, Section “ Conclusion and Future Directions ” concludes the study and outlines potential directions for future work. Literature review In this section, a survey on the detection of electricity-theft was highlighted. Each of the related study has been discussed with respect to the enlisted criterion: (a)- problem addressed and how the problem is addressed using their methodology; (b)- contribution of each research article; (c)- dataset with its size used for simulation; and (d)- results and analysis. In a recent study, a smart theft control framework in the logistics industry was proposed 36 . The authors identified vulnerable areas prone to theft and emphasized the integration of Digital Twin (DT) and Blockchain technologies to monitor operations in real time and support predictive decision-making. They implemented an IoT-based Blockchain logistics monitoring system to address theft vulnerabilities and performed real-time theft risk evaluation using a bi-directional convolutional neural network. The method demonstrated high performance, with prediction accuracy of 96.12%, specificity of 97.53%, and F-measure of 97.25%, though further testing in diverse operational environments was recommended. In another study, Khan et al. 37 proposed a supervised model for electricity theft detection (ETD). They addressed misclassification issues caused by imbalanced consumption data, overfitting, and high false positives by applying interpolation, the three-sigma rule, normalization, and the ADASYN algorithm for upsampling. Abnormal consumption patterns were detected using VGG-16, followed by a Firefly Algorithm-based XGBoost (FA-XGBoost) classifier. The model achieved an F1-score of 93.7%, precision of 92.6%, and recall of 97%. A collaborative system for smart grid security using Blockchain and wireless sensor networks was introduced in 38 . They highlighted vulnerabilities in SCADA systems, including DoS and False Data Injection Attacks (FDIA). Their framework used Proof-of-Authority Ethereum Blockchain and was evaluated on IEEE 14-, 30-, and 118-bus systems, applying statistical analyses (mean, standard deviation, skewness, kurtosis) to measure performance improvements. Lately, a machine-learning model for fraud detection using XGBoost and Random Forest, integrated with Blockchain features for secure and transparent operations 39 . Other studies have applied Blockchain in smart grids to reduce electricity theft costs. Distributed algorithms such as LUDP, OLUD, and LUD compute user honesty coefficients to detect fraudulent activity while preserving privacy 40 . A hybrid deep learning model (RNN-BiLSTM-CRF) was designed to detect electricity theft using multi-dimensional data. This model achieved 93.05% accuracy by combining sequence learning (RNN), bidirectional information capture (BiLSTM), and precise classification (CRF) 41 . Also, several studies have also combined Blockchain and LSTM-based anomaly detection for privacy-preserving electricity theft detection 42 . These approaches mitigate reliance on third parties, prevent tampering, and allow detection without revealing raw consumption data. Muzumdar et al. 43 developed a Blockchain-based energy theft detection system using smart contracts on Hyperledger Besu, achieving 98% detection accuracy, throughput exceeding 98.37 transactions/sec, and latency under 0.42 s in real-world tests. A Siamese Bi-LSTM hybrid model integrated with Blockchain for ETD, using cryptographic hash functions and BLS signatures to preserve data privacy was presented. Their hybrid model used RNN-GAN to generate synthetic theft data to address class imbalance, achieving high accuracy and low error rates while maintaining real-time power balancing 44 . Finally, a Blockchain-based model for real-time power balancing in decentralized renewable energy grids was presented 45 . By leveraging smart contracts, the system automates energy load balancing in real time, effectively responding to power fluctuations and demonstrating scalability for several hundred consumers. Based on the literature summarized in Table 1 , existing electricity theft detection approaches largely rely on either computationally intensive blockchain architectures or complex supervised deep learning models, each presenting limitations that hinder real-world deployment. Blockchain-based solutions often incur high computational and energy overhead, as prior works embed frequent model updates, continuous consensus operations, or large transaction payloads directly into the blockchain layer, resulting in poor scalability and increased latency. These issues arise from treating blockchain as a primary processing layer rather than a lightweight, immutable logging mechanism. Conversely, deep learning and hybrid neural models typically require balanced and labeled datasets, yet electricity theft data in real smart grids is inherently imbalanced and sparse, leading to performance degradation and overfitting in supervised architectures. Many studies also rely on simulated or laboratory-scale datasets, which fail to capture the temporal variability, noise, and irregular consumption patterns of operational grids. Hardware-based and IoT-centric solutions further complicate deployment due to maintenance costs and limited coverage. In contrast, the proposed Blockchain-supported LSTM-Autoencoder framework overcomes these challenges by employing unsupervised reconstruction-based learning to address class imbalance, using LSTM layers to model long-term temporal dependencies, and leveraging blockchain solely for tamper-proof event logging and transparent auditing. This approach avoids excessive on-chain computation while ensuring scalability, security, and real-time applicability. Table 1. Summary of existing state of art methods for electricity theft detection. Research study Method Dataset Contributions Limitations Results Khanet.al (2020) Blockchain, Reinforcement Learning, Federated Learning Proprietary smart meter dataset Apply Reinforcement detection and blockchain used for security High blockchain and learning overhead limits scalability and real-time deployment Accuracy 92%, Precision 90%,Recall 91%, F1-Score 90.5% Alanazi et.al (2024) IoT,NodeMCU, Arduino Real-time meter prototype data Low-cost hardware for theft monitoring Hardware-based monitoring lacks scalability and advanced data-driven detection Not directly applicable (Hardware monitoring system) Olivares et.al(2021) IoT,Blockchain, Digital Twin, Bi-Directional CNN Simulated load dataset Combine Digital twins with deep leaning High computational cost and weak long-term temporal modeling reduce efficiency Accuracy 94%, Precision 92%,Recall 93%, F1 Score 92.5% Calvagna et.al (2024) Blockchain, Smart Contracts Energy trading dataset Automatic power trading for smart contract Focuses on energy trading rather than real-time theft detection Not directly applicable (Power balancing service, not classification) Hasan et.al(2019) CNN, LSTM Smart grid consumption dataset High accuracy achieve for theft detection Model complexity and class imbalance sensitivity limit generalization Accuracy 96%, Precision 95%,Recall 94%, F1 Score 94.5% Open in a new tab Novel blockchain approach for electricity theft detection In this section, we present the proposed blockchain-based electricity theft detection framework for smart grid systems. The approach comprises several interlinked stages: (1) data preprocessing, (2) data balancing, (3) feature extraction, and (4) anomaly detection using LSTM-Autoencoder models. Together, these stages form an efficient and practical framework for detecting electricity theft. Figure 2 illustrates the overall system working and model design, which consists of two primary layers: the deep learning layer and the blockchain layer. Each component of the methodology is discussed in detail in the following subsections. Fig. 2. Open in a new tab Methodology for electricity theft detection using blockchain and deep learning models. Dataset collection Effective electricity theft detection relies on access to real-world consumption data. In this study, the dataset comprises electricity usage records from 16 different types of consumers, with hourly measurements recorded daily over a one-year period. To simulate fraudulent behavior, six distinct types of electricity theft were artificially introduced: Daytime reduction : Consumption is reduced during daytime hours by multiplying the original values by random factors between 0.1 and 0.8. Random zeroing : Consumption is set to zero for randomly selected time intervals. Random scaling : Each hourly consumption value is multiplied by a randomly generated number. Fractional mean : Consumption is set to a random fraction of the mean consumption. Fixed mean : Consumption is fixed at the mean value. Reversed pattern : The consumption pattern is reversed. A theft generator was developed to randomly introduce these fraudulent patterns into the dataset. The original consumption data were obtained from the Open Energy Data Initiative (OEDI), which maintains a comprehensive collection of high-quality energy research data curated by the U.S. Department of Energy 12 , 46 . Data pre-processing for electricity theft detection (ETD) Data preprocessing is a very significant step in the data analysis, ensuring the best performance of the model 47 . Further, it is critical for ensuring optimal model performance and reliable electricity theft detection. In this study, three fundamental preprocessing tasks are performed: handling missing values, outlier removal, and data normalization 37 . Missing consumption values, which may arise from meter malfunctions, communication failures, or transmission errors, are recovered using linear interpolation. Since electricity consumption data exhibit continuous temporal patterns, interpolation preserves time-series continuity by estimating missing values based on neighboring observations. For an observation xi : 1 This approach preserves the natural progression of consumption patterns and avoids abrupt discontinuities, maintaining the continuity of the time series. Next, outliers are addressed using the Three Sigma Rule 48 , where any value beyond two standard deviations from the mean is clipped to maintain consistency. Electricity consumption data can include anomalously large or small values due to meter faults, sudden load changes, or intentional interference. Such extreme values may distort the learning process, causing the model to interpret noise as significant behavior. To mitigate this, the Three Sigma Rule is applied: let μ represent the mean consumption, σ the standard deviation, and x an observed value. Any observation exceeding the threshold μ + 2σ is clipped, while all other values are retained. This statistically based method effectively eliminates extreme noise while preserving genuine variations in the data. By controlling outliers, the LSTM-Autoencoder focuses on true consumption patterns, stabilizing the learning process and reducing sensitivity to measurement errors. 2 Dataset mean and its standard deviation is thus computed using the model as givne in Eq. ( 3 ) below. Finally, the dataset is scaled into a uniform range [0, 1] using Min–Max Normalization to enhance convergence in deep learning models 37 : 3 In above equation, A denote the full consumption dataset, x norm the normalized value, and Min(A) and Max(A) the global minimum and maximum readings, respectively. Normalization scales the data to a standard range, which accelerates training, improves convergence, and prevents large-scale values from dominating the learning process. Together with handling missing values and outlier removal, these preprocessing steps refine the dataset, eliminate inconsistencies, and prepare it for robust learning in electricity theft detection. Data balancing In unbalanced datasets, the majority class dominates the learning process, causing the model to overlook patterns from the minority class. This leads to poor recognition of rare but critical events, such as electricity theft. Balancing the dataset ensures that the model learns representative features from all classes. The Customer Electricity Consumption dataset (Mendeley Data) contains far more regular usage records than theft instances, resulting in a highly skewed distribution that biases predictions toward normal consumption. To address this, we applied a synthetic oversampling technique to augment the minority class and achieve a more balanced dataset 49 . 4 where is a theft sample class, X neigℎbor is a k neariest neighbors is same class and is random scalar. It produces synthetic theft samples of various diversities to balance the dataset and enhance the capabilities of the LSTM autoencoder in identifying anomalies without being overly specific to them. Feature extraction Effective feature extraction is critical for electricity theft detection, as raw time-series consumption data alone may not adequately capture fraudulent behavior. Transforming raw data into meaningful features allows the model to learn behavioral, temporal, and statistical patterns associated with electricity misuse. First, time-based features, such as hour of day, day of week, and weekend or holiday indicators are incorporated to capture periodic consumption behavior, since legitimate electricity usage typically follows regular temporal patterns, and deviations may signal theft 46 . Second, statistical features, including mean, minimum, maximum, and standard deviation, provide insights into consumption variability and abnormal load behavior. Sudden peaks, drops, or irregular fluctuations serve as strong indicators of theft 19 , 20 . Thirdly, we added the most popular Load Factor (LF), which is defined to be: 5 In above equation, Pi i represent consumption during a given time period, and n denote the number of time intervals. Additionally, the load factor is computed to measure consumption stability over time, with low values typically indicating irregular or peaky usage that may suggest meter tampering or selective electricity theft. Difference-based features, capturing changes in consumption between consecutive time intervals, are also included to detect abrupt variations. These engineered features enhance the representational capacity of the LSTM-Autoencoder, improving its sensitivity to both short-term and long-term anomalies in smart grid environments. LSTM & autoencoders framework After preprocessing and feature extraction, the dataset is fed into the LSTM-Autoencoder model. This framework combines the strengths of Long Short-Term Memory (LSTM) networks, which effectively capture long-range temporal dependencies, with an Autoencoder, which learns compact latent representations of normal electricity consumption behavior. The approach is based on reconstruction-based anomaly detection: the Autoencoder is trained exclusively on normal consumption sequences to minimize reconstruction loss for legitimate behavior. In the event of theft, consumption patterns deviate from the learned manifold of normal behavior, resulting in a large reconstruction error E t , which serves as a strong indicator of anomalies. Smart meter data is recorded as a time-series sequence x t , reflecting daily or monthly consumption trends. Before model input, the data undergoes preprocessing, including linear interpolation for missing values, the Three Sigma Rule for outlier handling, and Min–Max normalization to stabilize training. In addition to raw consumption values, time-based features (hour, weekday, seasonal flags) and statistical descriptors (mean, max, standard deviation, load factor) are incorporated, enhancing the Autoencoder’s representational capacity and enabling the LSTM layers to capture periodicity, habitual usage cycles, and sudden behavioral changes. Within the LSTM-Autoencoder, the encoder compresses the temporal sequence into a lower-dimensional latent vector z, representing the extracted normal behavior. The decoder reconstructs the sequence from z. Formally, the encoder consists of a single LSTM layer with 64 hidden units, followed by two stacked LSTM layers with 128 hidden units each, compressing the input time series X = {× 1, × 2, … . xT , } , z ∈ R64 . To preserve temporal alignment, the latent vector z is repeated across all time steps using a Repeat Vector layer, preparing the input for the decoder. The decoder mirrors the encoder structure, with a single LSTM layer of 64 hidden units followed by a Time Distributed fully connected layer, which reconstructs the original feature space at each time step. A linear activation function, ReLU ( Rectified Linear Unit), is applied at the output to accurately reconstruct continuous electricity consumption values. Formally, the Autoencoder reconstruction process is defined as: 6 The model is optimized using the Mean Squared error (MSE) loss function: 7 In the proposed framework, MSE is selected as the reconstruction loss because it effectively penalizes deviations in continuous time-series data, making it particularly suitable for anomaly detection, where abnormal consumption patterns result in significantly higher reconstruction errors. The Autoencoder is trained exclusively on normal (non-theft) consumption sequences using the Adam optimizer, with early stopping applied to prevent overfitting by monitoring validation loss. During inference, the reconstruction error for each test sequence is computed and compared against a threshold derived from the upper percentile of normal reconstruction errors. Sequences exceeding this threshold are flagged as anomalous and forwarded to the blockchain layer for immutable logging and verification. The blockchain layer ensures data integrity and security while supporting automated decision-making through smart contracts. These contracts decentralize anomaly validation and provide tamper-proof records of detection events. Once the LSTM-Autoencoder identifies a suspicious pattern, the detection result is recorded as a blockchain transaction and submitted via the smart contract, enabling secure, transparent, and auditable electricity theft management. 8 where: ( x t ) is the input electricity consumption data at time t , ( y t ) is the prediction label (0 = normal, 1 = theft), ( E t ) is there construction error from the Autoencoder, time Stamp is the time of transaction logging. This function make sure that every theft detection is cryptographically logged with traceability, creating a permanent and tamper proof audit trail. Further, each transaction ( T x ) is encrypted with hashed and added more block in Blockchain: 9 Here, refers the current block, and Block n −1 is the one before block in Blockchain. A hash function provide security link with block and provide unique hash function. A small change in transaction data Has ℎ( T x ) will produce different hash function, helping and prevent any forgery or tampering. The Blockchain layer includes a smart contract-driven reward technique to promote honest consumption and discourage fraudulent behavior: 10 Compliant users may receive incentives, such as energy credits or discounts, while detected fraudsters may face automated service suspension, fines, or reporting to the utility. This mechanism encourages compliant behavior by rewarding adherence and deterring fraud, as illustrated in Figure 2 . The proposed model complete algorithm has been shown below in Algorithm 1. Experimental results and analysis This section presents a comparison of the proposed model’s results with those of related models evaluated on the same dataset. The experimental results demonstrate that the proposed model outperforms existing state-of-the-art approaches. To assess its effectiveness, the performance of the proposed model is compared against the baseline models listed below. Baseline for comparative analysis For comparative analysis, several baseline models have been selected, and each is described in detail below: Baseline 1 : Johncy et al. 50 proposed a blockchain-integrated Siamese Bi-LSTM model for electricity theft detection. The model leverages Generative Adversarial Networks (GANs) to address class imbalance and incorporates blockchain technology to secure energy transaction records. This approach enhances detection accuracy and precision while ensuring data privacy and integrity in smart grid systems. Baseline 2 : Hasan et al. 51 introduced a CNN-LSTM hybrid architecture for electricity theft detection. Convolutional layers perform automatic feature extraction, while LSTM layers capture temporal dependencies in electricity consumption patterns. The model also incorporates preprocessing techniques, including interpolation for missing values and synthetic data generation, to mitigate class imbalance issues. Simulation environment All simulations were implemented in Python, with the algorithms developed and trained using the TensorFlow deep learning library. The experiments were conducted on a system equipped with an Intel Core i5 processor and 8 GB of RAM , providing sufficient computational resources for model training and evaluation. Prior to model training, the dataset was carefully preprocessed using normalization and interpolation techniques to handle missing values and scale the data appropriately. The local Customer Electricity Consumption dataset served as the primary training dataset, enabling the proposed model to learn consumption patterns and detect anomalies effectively. This setup ensured a controlled and reproducible environment for evaluating the performance of the LSTM-Autoencoder framework in electricity theft detection. Performance metrics Electricity theft detection is formulated as a binary classification problem with highly imbalanced data. Evaluation metrics such as accuracy, recall, and F1-score are particularly relevant for assessing performance in imbalanced scenarios. The following metrics were employed to comprehensively evaluate the effectiveness of the proposed model. Accuracy : Accuracy is the percentage of the number of the cases that are correctly predicted by the model compared to the total number of cases. 11 Precision : The ratio of real positive cases to all predictions that are categorized as positive is known as precision, or positive predictive value. It shows positive predictions how many positive result achieve from dataset. 12 Recall : The True Positive Rate(TPR) is also called sensitivity and Recall ,Recall is a ratio of correctly predicted positive value to the total of positive classes within the dataset. 13 F1 Score : F1 score is statistic metric balance recall and precision to verify accuracy of classification model.it is define harmonic recall and precision, ensure balance of both measure especially when dealing with unbalance data. 14 ROC Curve : A plot of false positive rate (FPR) on true positive rate results in ROC curve which is used to evaluate how well a model performs on an unbalanced dataset .The scale of ROC curve is 0 to 1 and the model is said to perform effectively when the value of the ROC curve approaches 1. AUC (Area Under the Curve) is another value based on the ROC. 15 Result and analysis The experimental results of the proposed LSTM-Autoencoder model for electricity theft detection, using a time-series electricity consumption dataset, demonstrate a robust and well-structured anomaly detection framework. The process begins with data loading, cleaning, and normalization, where irrelevant columns are removed, missing values are handled, and features are scaled using Min–Max normalization. The theft label is encoded into a binary format for classification. The model architecture consists of an encoder LSTM layer, which compresses the temporal patterns of the input sequence into a latent representation, and a decoder LSTM layer, which reconstructs the original sequence from this compressed space. The objective is to learn normal consumption behavior by minimizing the mean squared reconstruction error, defined as the difference between the original and reconstructed sequences. The model is trained exclusively on normal (non-theft) data to capture baseline consumption patterns. As depicted in Fig. 3 below, Training is conducted for 20 epochs, during which the training and validation losses converge to approximately 6.3 × 10⁻ 5 , indicating that the model has effectively learned normal consumption behavior and can accurately reconstruct it for anomaly detection (Fig. 4 ). Fig. 3. Open in a new tab Training and validation loss over epochs. Fig. 4. Open in a new tab ROC curve of model performance. After training, the reconstruction error is computed for each test instance, and a threshold, commonly set at the 95th percentile, is used to identify anomalies. However, values exceeding this threshold are classified as potential electricity theft. With the full implementation of the proposed model, including data balancing using SMOTE and optimized hyperparameters, the model successfully detected theft cases from the dataset, demonstrating its strong ability to distinguish between normal and fraudulent consumption patterns. The model achieved an overall accuracy of approximately 95%, recall of 95%, and F1-score of 97% for theft classes. The confusion matrix further, as shown in Fig. 5 , confirms that the model accurately identifies theft cases while correctly classifying normal instances. Fig. 5. Open in a new tab Confusion matrix of LSTM-Autoencoder. Although the loss curve exhibits smooth convergence and the ROC curve provides insight into class separability, the absence of sufficient positive samples in the test set can lead to undefined or misleading precision, recall, and F1 scores, it has been shown in above Fig. 4 , which shows the significant model performance. This underscores the critical importance of maintaining a balanced dataset, particularly during evaluation, to ensure meaningful performance metrics. Techniques such as SMOTE effectively address class imbalance, allowing the model to learn representative features of theft events. Additionally, integrating advanced architectures, including attention mechanisms or hybrid deep learning pipelines, can further enhance performance in imbalanced scenarios. In summary, while the LSTM-Autoencoder effectively captures normal electricity usage patterns and detects anomalies, its real-world deployment for electricity theft detection requires careful dataset balancing, threshold tuning, and robust evaluation with adequate representation of anomalous behavior to ensure reliable and accurate detection. Baseline-1 results: siamese Bi-LSTM model The blockchain-integrated Siamese Bi-LSTM framework proposed by Johncy et al. 50 demonstrates robust performance in electricity theft detection. By leveraging BLS-based privacy-preserving blockchain, the framework ensures data integrity and confidentiality, while the Siamese Bi-LSTM network effectively differentiates between normal and fraudulent electricity consumption patterns. The framework uses GAN-based data augmentation to address class imbalance, further improving detection accuracy. The experimental results demonstrate an accuracy of 98.2%, precision of 97.51%, recall of 99.87%, and an average transaction time of 0.8 s, confirming the framework’s effectiveness, efficiency, and scalability for real-time electricity theft detection. These observed results have been shown in Figs. 6 and 7 . Fig. 6. Open in a new tab Recall of models. Fig. 7. Open in a new tab Precision of models. Baseline-2 results: CNN-LSTM model The CNN-LSTM model proposed by Hasan et al. 51 integrates spatial feature extraction with temporal modeling to detect electricity theft. To mitigate class imbalance, synthetic oversampling techniques are applied during preprocessing. Experimental evaluation demonstrates that the model achieves an accuracy of 93.92%, precision of 94.12%, and an F1-score of 93.83%, indicating that the CNN-LSTM architecture effectively identifies both legitimate and fraudulent electricity consumption patterns. In addition to the previously discussed studies, a detailed comparative analysis was conducted to evaluate the effectiveness of the proposed Blockchain-integrated LSTM-Autoencoder framework against prominent baseline models. The proposed model achieves an accuracy of 95.0%, recall of 95.0%, and an F1-score of 97.0%, as given in Fig. 8 above and it is demonstrating its strong capability to detect electricity theft. Fig. 8. Open in a new tab Result of CNN-LSTM. The first baseline, the CNN-LSTM hybrid model, utilizes convolutional layers for spatial feature extraction and LSTM units for capturing temporal dependencies. This model achieved an accuracy of 94.12%, recall of 93.92%, and an F1-score of 93.83%, reflecting its effectiveness in identifying both normal and fraudulent consumption patterns. The second baseline, the Siamese Bi-LSTM model integrated with BLS privacy blockchain, combines privacy-preserving blockchain with a Siamese Bi-LSTM network for secure and accurate electricity theft detection. Experimental results indicate superior performance, with 98.2% accuracy, 97.51% precision, and 99.87% recall, highlighting its strength in ensuring data privacy while maintaining reliable theft detection. These results have been presented in Table 2 below. Table 2. Comparison table of models. Performance metrics Proposed model CNN-LSTM Siamese Bi-LSTM model Accuracy 95.00% 94.12% 98.2% Recall 95.00% 93.92% 99.87% F1-Score 97.00% 93.83% 97.73% Open in a new tab Overall, the comparative evaluation confirms, which have been shown in Fig. 9 , that the proposed Blockchain-integrated LSTM-Autoencoder achieves competitive performance, balancing detection accuracy, robustness, and scalability, while leveraging blockchain for secure, transparent, and auditable operations. Fig. 9. Open in a new tab Performance compassion of models. The high recall value demonstrates a low false negative rate, ensuring that fraudulent electricity consumption is reliably detected. Furthermore, the integration of blockchain technology enhances data security and transparency by immutably recording all detected anomalies through smart contracts. By combining advanced deep learning with decentralized security mechanisms, the proposed framework offers a scalable, trustworthy, and intelligent solution for electricity theft detection in modern smart grid environments. Conclusion and future directions This study proposes a novel electricity theft detection framework for smart grids by integrating LSTM-Autoencoder deep learning with blockchain technology. The framework addresses critical challenges such as class imbalance, temporal dependency modeling, data security, and transparency. By leveraging reconstruction-based anomaly detection, the LSTM-Autoencoder effectively captures normal consumption patterns and identifies fraudulent behavior through elevated reconstruction errors. Experimental results demonstrate an accuracy of 95% with strong ROC-AUC, confirming reliable detection performance. The blockchain layer ensures immutable logging of anomalies and supports automated decision-making via smart contracts, enhancing trust, transparency, and accountability. This decentralized design incentivizes compliant behavior while deterring fraud, offering a scalable, privacy-preserving, and reliable solution for reducing non-technical losses in smart grids. Future work may focus on multimodal data fusion, hybrid or reinforcement learning frameworks, and integration with AMI and SCADA systems under blockchain governance, further enhancing detection accuracy, system flexibility, and security in complex, real-world smart grid environments. Supplementary Information Supplementary Information. (17.6KB, docx) Acknowledgements The authors extend their appreciation to the Deanship of Scientific Research at Northern Border University, Arar, KSA for funding this research work through the project number “NBU-FFR-2026-2990-01” Author contributions Fadia Bibi and Saif ur rehman was involved in the main concept, writing and drafting the article. Vincent Karovič and Sarfraz Biib helped to finalize the article, proof read, funding and data collection. Ahmad Alshammari provided resourcesfor accomplishing the simulation tasks involved in the study. Kamran aziz was involved in the investigation and analysis. Data availability Data is available on https://data.mendeley.com/datasets/c3c7329tjj/1 [Accessed on 25 September 2025]. Declarations Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Contributor Information Saif Ur Rehman, Email: [email protected]. Vincent Karovič, Email: [email protected]. References 1. Wang, J., Cui, H., Hao, H., Fig 4, H., & Bao, L. (2018). Research on digital power grid information integration solution. In MATEC Web of Conferences (Vol. 173, p. 03026). EDP Sciences 2. Tariq, M., Ali, M., Naeem, F. & Poor, H. V. Vulnerability assessment of 6Genabled smart grid cyber–physical systems. IEEE Internet Things J. 8 (7), 5468–5475 (2020). [ Google Scholar ] 3. Adejumo, A. P., & Ogburie, C. (2025). Blockchain for Fraud Prevention: Transforming Accounting and Finance. Futurity Proceedings, 2. 4. Barai, G. R., Krishnan, S., & Venkatesh, B. (2015, October). Smart metering and functionalities of smart meters in smart grid-a review. In 2015 IEEE Electrical Power and Energy Conference (EPEC) (pp. 138–145). IEEE. 5. Mohajeri, M., Ghassemi, A. & Gulliver, T. A. Fast big data analytics for smart meter data. IEEE Open J. Commun. Soc. 1 , 1864–1871 (2020). [ Google Scholar ] 6. McDaniel, P. & McLaughlin, S. Security and privacy challenges in the smart grid. IEEE Secur. Priv. 7 (3), 75–77 (2009). [ Google Scholar ] 7. Ekanayake, J. B., Jenkins, N., Liyanage, K. M., Wu, J. & Yokoyama, A. Smart grid: technology and applications (John Wiley & Sons, 2012). [ Google Scholar ] 8. Ahmad, T., Chen, H., Wang, J. & Guo, Y. Review of various modeling techniques for the detection of electricity theft in smart grid environment. Renew. Sustain. Energy Rev. 82 , 2916–2933 (2018). [ Google Scholar ] 9. Ericsson, G. N. Cyber security and power system communication—essential parts of a smart grid infrastructure. IEEE Trans. Power Deliv. 25 (3), 1501–1507 (2010). [ Google Scholar ] 10. Yao, D. et al. Energy theft detection with energy privacy preservation in the smart grid. IEEE Internet Things J. 6 (5), 7659–7669 (2019). [ Google Scholar ] 11. Aldegheishem, A. et al. Towards sustainable energy efficiency with intelligent electricity theft detection in smart grids emphasising enhanced neural networks. IEEE Access 9 , 25036–25061 (2021). [ Google Scholar ] 12. Nawaz, A., Ali, T., Mustafa, G., Rehman, S. U. & Rashid, M. R. A novel technique for detecting electricity theft in secure smart grids using CNN and XG-boost. Intell. Syst. Appl. 17 , 200168 (2023). [ Google Scholar ] 13. Jameel, S. (2018). An optimal feature selection method using a modified wrapper-based ant colony optimisation. Journal of the National Science Foundation of Sri Lanka, 46(2). 14. Khan, S., Kifayat, K., Kashif Bashir, A., Gurtov, A. & Hassan, M. Intelligent intrusion detection system in smart grid using computational intelligence and machine learning. Trans. Emerg. Telecommun. Technol. 32 (6), e4062 (2021). [ Google Scholar ] 15. Punmiya, R. & Choe, S. Energy theft detection using gradient boosting theft detector with feature engineering-based preprocessing. IEEE Trans. Smart Grid 10 (2), 2326–2329 (2019). [ Google Scholar ] 16. Lazaropoulos, A. G. Detection of energy theft in overhead low-voltage power grids-The hook style energy theft in the smart grid era. Trends Renew. Energy 5 (1), 12–46 (2018). [ Google Scholar ] 17. Madhure, R. U., Raman, R., & Singh, S. K. (2020, July). Cnn-lstm based electricity theft detector in advanced metering infrastructure. In 2020 11th International Conference on Computing, Communication and Networking Technologies (ICCCNT) (pp. 1–6). IEEE. 18. Ferrag, M. A. et al. Blockchain technologies for the internet of things: Research issues and challenges. IEEE Internet Things J. 6 (2), 2188–2204 (2018). [ Google Scholar ] 19. Himeur, Y., Ghanem, K., Alsalemi, A., Bensaali, F., & Amira, A. (2010). Anomaly detection of energy consumption in buildings: A review, current trends and new perspectives. arXiv 2020. arXiv preprint arXiv:2010.04560 . 20. Adil, M. et al. LSTM and bat-based RUSBoost approach for electricity theft detection. Appl. Sci. 10 (12), 4378 (2020). [ Google Scholar ] 21. Zheng, Z., Yang, Y., Niu, X., Dai, H. N. & Zhou, Y. Wide and deep convolutional neural networks for electricity-theft detection to secure smart grids. IEEE Trans. Ind. Inform. 14 (4), 1606–1615 (2017). [ Google Scholar ] 22. Abdulqadder, I. H., Aziz, I. T. & Flaih, F. M. Robust electricity theft detection in smart grids using machine learning and secure techniques. Int. J. Intel. Eng. Syst. 18 (1), 1021 (2025). [ Google Scholar ] 23. Yao, A., Jiang, F., Li, X., Dong, C., Xu, J., Xu, Y., ... & Liu, X. (2021, October). A novel security framework for edge computing based uav delivery system. In 2021 IEEE 20th International Conference on Trust, Security and Privacy in Computing and Communications (TrustCom) (pp. 1031–1038). IEEE. 24. Amine Ferrag, M., Derdour, M., Mukherjee, M., Derhab, A., Maglaras, L., & Janicke, H. (2018). Blockchain technologies for the internet of things: research issues and challenges. arXiv e-prints, arXiv-1806. 25. Faheem, M. & Al-Khasawneh, M. A. Multilayer cyberattacks identification and classification using machine learning in internet of blockchain (IoBC)-based energy networks. Data Brief. 54 , 110461 (2024). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 26. Lu, K. D., Wu, Z. G. & Huang, T. Differential evolution-based three stage dynamic cyber-attack of cyber-physical power systems. IEEE/ASME Trans. Mechatron. 28 (2), 1137–1148 (2022). [ Google Scholar ] 27. Bandim, C. J., Alves, J. E. R., Pinto, A. V., Souza, F. C., Loureiro, M. R., Magalhaes, C. A., & Galvez-Durand, F. (2003, September). Identification of energy theft and tampered meters using a central observer meter: a mathematical approach. In 2003 IEEE PES Transmission and Distribution Conference and Exposition (IEEE Cat. No. 03CH37495) (Vol. 1, pp. 163–168). IEEE. 28. Ibrahim, N. M., Al-Janabi, S. T. F. & Al-Khateeb, B. Electricity-theft detection in smart grids based on deep learning. Bull. Electr. Eng. Inform. 10 (4), 2285–2292 (2021). [ Google Scholar ] 29. Gunduz, M. Z. & Das, R. Smart grid security: An effective hybrid CNNbased approach for detecting energy theft using consumption patterns. Sensors 24 (4), 1148 (2024). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 30. Grid, N. S. (2010). Guidelines for smart grid cyber security: Vol. 2, privacy and the smart grid. Guideline, Aug, 6. 31. Dong, C., Jiang, F., Li, X., Yao, A., Li, G., & Liu, X. (2021, May). A blockchain-aided self-sovereign identity framework for edge-based UAV delivery system. In 2021 IEEE/ACM 21st International Symposium on Cluster, Cloud and Internet Computing (CCGrid) (pp. 622–624). IEEE. 32. Dong, C., Pal, S., Chen, S., Jiang, F. & Liu, X. A privacy-aware task distribution architecture for UAV communications system using blockchain. IEEE Internet Things J. 12 (9), 11233–11243 (2025). [ Google Scholar ] 33. Yao, A. et al. FedShufde: A privacy preserving framework of federated learning for edge-based smart UAV delivery system. Future Generat. Computer Syst. 166 , 107706 (2025). [ Google Scholar ] 34. Yao, A. et al. A privacy-preserving location data collection framework for intelligent systems in edge computing. Ad Hoc Netw. 161 , 103532 (2024). [ Google Scholar ] 35. Johncy, G., Shaji, R. S., Angelin Monisha Sharean, T. M. & Hubert, U. Enhancing smart grid security using BLS privacy blockcha in with siamese BiLSTM for electricity theft detection. Trans. Emerg. Telecommun. Technol. 36 (1), e70033 (2025). [ Google Scholar ] 36. Dhasaratha, C., Hasan, M. K., Islam, S., Khapre, S., Abdullah, S., Ghazal, T. M., ... & Akhtaruzzaman, M. (2024). Data privacy model using blockchain reinforcement federated learning approach for scalable internet of medical things. CAAI Transactions on Intelligence Technology. 37. Khan, Z. A. et al. Electricity theft detection using supervised learning techniques on smart meter data. Sustainability 12 (19), 8023 (2020). [ Google Scholar ] 38. Almasabi, S. et al. Securing smart grid data with blockchain and wireless sensor networks: A collaborative approach. IEEE Access 12 , 19181–19198 (2024). [ Google Scholar ] 39. Ashfaq, T. et al. A machine learning and blockchain based efficient fraud detection mechanism. Sensors 22 (19), 7162 (2022). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 40. Din, J., Su, H., Ali, S. & Salman, M. Research on blockchain-enabled smart grid for anti-theft electricity securing peer-to-peer transactions in modern grids. Sensors 24 (5), 1668 (2024). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 41. Khalid, A., Mustafa, G., Rana, M. R. R., Alshahrani, S. M. & Alymani, M. RNN-BiLSTM-CRF based amalgamated deep learning model for electricity theft detection to secure smart grids. PeerJ Comput. Sci. 10 , e1872 (2024). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 42. Zhao, Z., Liu, Y., Zeng, Z., Chen, Z. & Zhou, H. Privacy-preserving electricity theft detection based on blockchain. IEEE Trans. Smart Grid 14 (5), 4047–4059 (2023). [ Google Scholar ] 43. Muzumdar, A., Modi, C. & Vyjayanthi, C. Designing a blockchain-enabled privacy-preserving energy theft detection system for smart grid neighborhood area network. Electr. Power Syst. Res. 207 , 107884 (2022). [ Google Scholar ] 44. Johncy, G., Shaji, R. S., & Angelin Monisha Sharean, T. M. (2023). A Novel Deep Learning Model for Privacy based Blockchain with Electricity Theft Detection in Smart Grid. 45. Calvagna, A., Marotta, G., Pappalardo, G. & Tramontana, E. A blockchainbased real-time power balancing service for trustless renewable energy grids. Future Internet 16 (5), 149 (2024). [ Google Scholar ] 46. Haq, E. U., Pei, C., Zhang, R., Jianjun, H. & Ahmad, F. Electricity-theft detection for smart grid security using smart meter data: A deep-CNN based approach. Energy Rep. 9 , 634–643 (2023). [ Google Scholar ] 47. Kanwal, B. et al. Opinion mining from online travel reviews: An exploratory investigation on Pakistan major online travel services using natural language processing. IEEE Access 11 , 29934–29945 (2023). [ Google Scholar ] 48. Ayub, N., Ali, U., Mustafa, K., Mohsin, S. M. & Aslam, S. Predictive data analytics for electricity fraud detection using tuned CNN ensembler in smart grid. Forecasting 4 (4), 936–948 (2022). [ Google Scholar ] 49. Johncy, G., Shaji, R. S., Angelin Monisha Sharean, T. M., & Hubert, U. (2024). Enhancing smart grid security using BLS privacy blockchain with siamese Bi-LSTM for electricity theft detection. Department of Information Technology, st. Xavier’s Catholic College of Engineering, Nagercoil, India. 50. Ferrag, M. A. & Maglaras, L. DeepCoin: A novel deep learning and blockchain-based energy exchange framework for smart grids. IEEE Trans. Eng. Manag. 67 (4), 1285–1297 (2019). [ Google Scholar ] 51. Hasan, M. N., Toma, R. N., Nahid, A. A., Islam, M. M. & Kim, J. M. Electricity theft detection in smart grid systems: A CNN-LSTM based approach. Energies 12 (17), 3310 (2019). [ Google Scholar ] 52. Kilimci, Z. H. & Akyokus, S. Deep learning‐and word embedding‐based heterogeneous classifier ensembles for text classification. Complexity 2018 (1), 7130146 (2018). [ Google Scholar ] 53. Quinn, E. L. (2009). Privacy and the new energy infrastructure. Available at SSRN 1370731. 54. Alanazi, A., Alqahtani, A., Alsubai, S. & Bhatia, M. IoT-inspired smart theft control framework for logistic industry. IEEE Internet Things J. 10.1109/jiot.2024.3445884 (2024). [ Google Scholar ] 55. Olivares-Rojas, J. C. et al. A transactive energy model for smart metering systems using blockchain. CSEE J. Power Energy Syst. 7 (5), 943–953 (2021). [ Google Scholar ] 56. Mazhar, T. et al. Analysis of cyber security attacks and its solutions for the smart grid using machine learning and blockchain methods. Future Internet 15 (2), 83 (2023). [ Google Scholar ] 57. Lu, K. D. & Wu, Z. G. Constrained-differential-evolution-based stealthy sparse cyber-attack and countermeasure in an AC smart grid. IEEE Trans. Ind. Inform. 18 (8), 5275–5285 (2021). [ Google Scholar ] Associated Data This section collects any data citations, data availability statements, or supplementary materials included in this article. Supplementary Materials Supplementary Information. (17.6KB, docx) Data Availability Statement Data is available on https://data.mendeley.com/datasets/c3c7329tjj/1 [Accessed on 25 September 2025]. Articles from Scientific Reports are provided here courtesy of Nature Publishing Group ACTIONS View on publisher site PDF (2.3 MB) Cite Collections Permalink PERMALINK Copy RESOURCES Similar articles Cited by other articles Links to NCBI Databases Cite Copy Download .nbib .nbib Format: AMA APA MLA NLM Add to Collections Create a new collection Add to an existing collection Name your collection * Choose a collection Unable to load your collection due to an error Please try again Add Cancel Follow NCBI NCBI on X (formerly known as Twitter) NCBI on Facebook NCBI on LinkedIn NCBI on GitHub NCBI RSS feed Connect with NLM NLM on X (formerly known as Twitter) NLM on Facebook NLM on YouTube National Library of Medicine 8600 Rockville Pike Bethesda, MD 20894 Web Policies FOIA HHS Vulnerability Disclosure Help Accessibility Careers NLM NIH HHS USA.gov Back to Top