Zhirui Luo
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Zhirui Luo has a total of 4 co-authors. Here are the top co-authors by number of shared publications:
- Jun Zheng
Department of Computer Science and Engineering, New Mexico Institute of Mining and Technology, Socorro, NM, USA
11 shared publications - Qingqing Li
Department of Computer and Information Sciences, Towson University, Towson, MD, USA
11 shared publications - Ruobin Qi
Department of Computer Science, California State University, Northridge, USA
8 shared publications - Sihua Shao
Colorado School of Mines
2 shared publications
Latest Publications
- Open Access
- 16 March 2026
- Published by Springer Nature in Neural Computing and Applications
- Vol. 38 (6) , 172
MOGSM-Net: Multi-objective genetically searched multi-label deep networks for non-intrusive load monitoring
Deep learning has recently emerged to address the non-intrusive load monitoring (NILM) problem. Existing deep learning methods for NILM often rely on manually designed models that struggle to efficiently capture temporal and spatial patterns in smart meter data. These models also tend to overlook computational constraints. This paper introduces MOGSM-Net, a multi-label deep learning model for NILM designed automatically through neural architecture search (NAS). MOGSM-Net combines convolutional neural networks (CNNs), long short-term memory (LSTM) networks, and attention mechanisms in a lightweight hybrid architecture. It employs a multi-label classification head with an asymmetric loss function to handle the highly imbalanced data inherent to the NILM problem. To discover high-performance and lightweight MOGSM-Net architectures, we developed a multi-objective genetic search algorithm. Performance evaluation using two publicly available NILM datasets shows that MOGSM-Net outperforms existing state-of-the-art methods while remaining computationally efficient, making it a promising solution for NILM.
- 16 March 2026
- Published by Springer Nature in Neural Computing and Applications
- Vol. 38 (6) , 172
Open AccessMOGSM-Net: Multi-objective genetically searched multi-label deep networks for non-intrusive load monitoring
Deep learning has recently emerged to address the non-intrusive load monitoring (NILM) problem. Existing deep learning methods for NILM often rely on manually designed models that struggle to efficiently capture temporal and spatial patterns in smart meter data. These models also tend to overlook computational constraints. This paper introduces MOGSM-Net, a multi-label deep learning model for NILM designed automatically through neural architecture search (NAS). MOGSM-Net combines convolutional neural networks (CNNs), long short-term memory (LSTM) networks, and attention mechanisms in a lightweight hybrid architecture. It employs a multi-label classification head with an asymmetric loss function to handle the highly imbalanced data inherent to the NILM problem. To discover high-performance and lightweight MOGSM-Net architectures, we developed a multi-objective genetic search algorithm. Performance evaluation using two publicly available NILM datasets shows that MOGSM-Net outperforms existing state-of-the-art methods while remaining computationally efficient, making it a promising solution for NILM.
- Open Access
- 10 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 210391-210405
A Data-Driven Unsupervised Ensemble Method for Electricity Theft Detection in AMI Systems
Smart meter data generated by Advanced Metering Infrastructure (AMI) systems is vulnerable to false data injection attacks (FDIAs), which are often carried out by electricity thieves to lower their utility bills. Unsupervised electricity theft detection (ETD) methods aim to identify fraudulent users using only unlabeled data. However, existing unsupervised approaches struggle to effectively detect both power consumption reduction attacks and load profile shifting attacks. In this paper, we propose a novel unsupervised ensemble method capable of detecting both types of attacks. The proposed method integrates three individual detection algorithms: two unsupervised anomaly detection algorithms, K-nearest neighbors (KNN) and Gaussian mixture model (GMM), and a correlation-based detection algorithm using mutual information (MI). We apply a weighted combination strategy to aggregate the user anomaly degrees assessed by these algorithms, with weights determined based on their detection capabilities using K-means clustering. We conducted a performance evaluation using smart meter datasets from both business and residential users. Experimental results show that our ensemble approach significantly outperforms state-of-the-art baselines, demonstrating its effectiveness as a robust solution for unsupervised ETD.
- 10 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 210391-210405
Open AccessA Data-Driven Unsupervised Ensemble Method for Electricity Theft Detection in AMI Systems
Smart meter data generated by Advanced Metering Infrastructure (AMI) systems is vulnerable to false data injection attacks (FDIAs), which are often carried out by electricity thieves to lower their utility bills. Unsupervised electricity theft detection (ETD) methods aim to identify fraudulent users using only unlabeled data. However, existing unsupervised approaches struggle to effectively detect both power consumption reduction attacks and load profile shifting attacks. In this paper, we propose a novel unsupervised ensemble method capable of detecting both types of attacks. The proposed method integrates three individual detection algorithms: two unsupervised anomaly detection algorithms, K-nearest neighbors (KNN) and Gaussian mixture model (GMM), and a correlation-based detection algorithm using mutual information (MI). We apply a weighted combination strategy to aggregate the user anomaly degrees assessed by these algorithms, with weights determined based on their detection capabilities using K-means clustering. We conducted a performance evaluation using smart meter datasets from both business and residential users. Experimental results show that our ensemble approach significantly outperforms state-of-the-art baselines, demonstrating its effectiveness as a robust solution for unsupervised ETD.
- Open Access
Designing Channel Attention Fully Convolutional Networks with Neural Architecture Search for Customer Socio-Demographic Information Identification Using Smart Meter Data
Background: Accurately identifying the socio-demographic information of customers is crucial for utilities. It enables them to efficiently deliver personalized energy services and manage distribution networks. In recent years, machine learning-based data-driven methods have gained popularity compared to traditional survey-based approaches, owing to their time and cost efficiency, as well as the availability of a large amount of high-frequency smart meter data. Methods: In this paper, we propose a new method that harnesses the power of neural architecture search to automatically design deep neural network architectures tailored for identifying various socio-demographic information of customers using smart meter data. We designed a search space based on a novel channel attention fully convolutional network architecture. Furthermore, we developed a search algorithm based on Bayesian optimization to effectively explore the space and identify high-performing architectures. Results: The performance of the proposed method was evaluated and compared with a set of machine learning and deep learning baseline methods using a smart meter dataset widely used in this research area. Our results show that the deep neural network architectures designed automatically by our proposed method significantly outperform all baseline methods in addressing the socio-demographic questions investigated in our study.
Open AccessDesigning Channel Attention Fully Convolutional Networks with Neural Architecture Search for Customer Socio-Demographic Information Identification Using Smart Meter Data
Background: Accurately identifying the socio-demographic information of customers is crucial for utilities. It enables them to efficiently deliver personalized energy services and manage distribution networks. In recent years, machine learning-based data-driven methods have gained popularity compared to traditional survey-based approaches, owing to their time and cost efficiency, as well as the availability of a large amount of high-frequency smart meter data. Methods: In this paper, we propose a new method that harnesses the power of neural architecture search to automatically design deep neural network architectures tailored for identifying various socio-demographic information of customers using smart meter data. We designed a search space based on a novel channel attention fully convolutional network architecture. Furthermore, we developed a search algorithm based on Bayesian optimization to effectively explore the space and identify high-performing architectures. Results: The performance of the proposed method was evaluated and compared with a set of machine learning and deep learning baseline methods using a smart meter dataset widely used in this research area. Our results show that the deep neural network architectures designed automatically by our proposed method significantly outperform all baseline methods in addressing the socio-demographic questions investigated in our study.
- 13 May 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 73 (00189456) , 1-11
Automatic Searching of Lightweight and High-Performing CNN Architectures for EEG-Based Driving Fatigue Detection
The increasing number of vehicles has led to a rise in traffic accidents, with fatigued driving being a major contributing factor. Bio-electrical signals, particularly electroencephalograms (EEG), have emerged as a promising avenue for detecting driving fatigue. EEG signals can provide valuable insights into a person’s brain activity and state of alertness. However, the complexity of EEG signals and the need for real-time detection pose significant challenges for traditional machine learning algorithms, leading to the growing popularity of deep learning in this domain. The objective of this article is to design lightweight and high-performing convolutional neural network (CNN) models for detecting driving fatigue using multichannel EEG signals. These models are intended to be deployed on resource-limited devices in intelligent vehicles, enabling timely alerts for fatigued driving. Rather than manually designing the deep neural network (DNN) architecture, we adopted the neural architecture search (NAS) approach to automate the architecture design process, considering both detection performance and computational cost. To evaluate the effectiveness of our approach, we conducted experiments using two publicly available EEG datasets widely used in driving fatigue detection studies. The performance of our NAS-derived model, named FD-LiteNet, was compared with a set of state-of-the-art baseline CNN models manually designed for EEG signal analysis. The results demonstrate that FD-LiteNet achieves significantly higher detection accuracy than all baseline models with a lower computational cost. Furthermore, our findings highlight the exceptional generalization capability of FD-LiteNet, as it can be fine-tuned with a small number of new samples to adapt to new datasets.
- 13 May 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 73 (00189456) , 1-11
Automatic Searching of Lightweight and High-Performing CNN Architectures for EEG-Based Driving Fatigue Detection
The increasing number of vehicles has led to a rise in traffic accidents, with fatigued driving being a major contributing factor. Bio-electrical signals, particularly electroencephalograms (EEG), have emerged as a promising avenue for detecting driving fatigue. EEG signals can provide valuable insights into a person’s brain activity and state of alertness. However, the complexity of EEG signals and the need for real-time detection pose significant challenges for traditional machine learning algorithms, leading to the growing popularity of deep learning in this domain. The objective of this article is to design lightweight and high-performing convolutional neural network (CNN) models for detecting driving fatigue using multichannel EEG signals. These models are intended to be deployed on resource-limited devices in intelligent vehicles, enabling timely alerts for fatigued driving. Rather than manually designing the deep neural network (DNN) architecture, we adopted the neural architecture search (NAS) approach to automate the architecture design process, considering both detection performance and computational cost. To evaluate the effectiveness of our approach, we conducted experiments using two publicly available EEG datasets widely used in driving fatigue detection studies. The performance of our NAS-derived model, named FD-LiteNet, was compared with a set of state-of-the-art baseline CNN models manually designed for EEG signal analysis. The results demonstrate that FD-LiteNet achieves significantly higher detection accuracy than all baseline models with a lower computational cost. Furthermore, our findings highlight the exceptional generalization capability of FD-LiteNet, as it can be fine-tuned with a small number of new samples to adapt to new datasets.
- 16 November 2023
- Published by Elsevier in Sustainable Energy, Grids and Networks
- Vol. 36
Deep semi-supervised electricity theft detection in AMI for sustainable and secure smart grids
- 16 November 2023
- Published by Elsevier in Sustainable Energy, Grids and Networks
- Vol. 36
Deep semi-supervised electricity theft detection in AMI for sustainable and secure smart grids
- Open Access
- 5 September 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 11, 96797-96807
DeepTPA-Net: A Deep Triple Attention Network for sEMG-Based Hand Gesture Recognition
The use of hand gestures for human-computer interaction (HCI) has gained popularity due to its ability to provide natural and intuitive communication in human dialogues. Hand gesture recognition (HGR) using surface electromyography (sEMG) signals is more reliable and user-friendly than traditional computer vision-based methods. This study proposes a deep network named DeepTPA-Net that utilizes multi-channel sEMG signals to recognize hand gestures. DeepTPA-Net employs a ResNet50 network as an automated feature extractor and a novel triple attention (3Attn) block that connects spatial, temporal, and channel attention modules in parallel to signify important features for HGR. We evaluated the performance of DeepTPA-Net using five publicly available benchmark sEMG hand gesture datasets, including CapgMyo DB-a, Csl-hdemg, NinaPro DB1, NinaPro DB2, and SeNic. The effectiveness of the proposed 3Attn block for HGR is demonstrated through a performance comparison with other attention mechanisms. We compared the performance of DeepTPA-Net with various baseline models, including its variations and other existing methods. The results show that DeepTPA-Net significantly outperforms the baseline models for all five benchmark datasets, indicating the superiority of DeepTPA-Net for sEMG-based HGR.
- 5 September 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 11, 96797-96807
Open AccessDeepTPA-Net: A Deep Triple Attention Network for sEMG-Based Hand Gesture Recognition
The use of hand gestures for human-computer interaction (HCI) has gained popularity due to its ability to provide natural and intuitive communication in human dialogues. Hand gesture recognition (HGR) using surface electromyography (sEMG) signals is more reliable and user-friendly than traditional computer vision-based methods. This study proposes a deep network named DeepTPA-Net that utilizes multi-channel sEMG signals to recognize hand gestures. DeepTPA-Net employs a ResNet50 network as an automated feature extractor and a novel triple attention (3Attn) block that connects spatial, temporal, and channel attention modules in parallel to signify important features for HGR. We evaluated the performance of DeepTPA-Net using five publicly available benchmark sEMG hand gesture datasets, including CapgMyo DB-a, Csl-hdemg, NinaPro DB1, NinaPro DB2, and SeNic. The effectiveness of the proposed 3Attn block for HGR is demonstrated through a performance comparison with other attention mechanisms. We compared the performance of DeepTPA-Net with various baseline models, including its variations and other existing methods. The results show that DeepTPA-Net significantly outperforms the baseline models for all five benchmark datasets, indicating the superiority of DeepTPA-Net for sEMG-based HGR.
- 31 March 2023
- Published by Springer Nature
- p. 152-164
ABODE-Net: An Attention-based Deep Learning Model for Non-intrusive Building Occupancy Detection Using Smart Meter Data
- 31 March 2023
- Published by Springer Nature
- p. 152-164
ABODE-Net: An Attention-based Deep Learning Model for Non-intrusive Building Occupancy Detection Using Smart Meter Data
- 11 July 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 71 (00189456) , 1-10
A Novel Unsupervised Data-Driven Method for Electricity Theft Detection in AMI Using Observer Meters
The smart meter data of the advanced metering infrastructure (AMI) can be tampered by electricity thieves with advanced digital instruments or cyber attacks to reduce their electricity bills, which causes devastating financial losses to utilities. A novel unsupervised data-driven method for electricity theft detection in AMI is proposed in this article. The method incorporates observer meter data, wavelet-based feature extraction, and fuzzy c-means (FCM) clustering. A new anomaly score is developed based on the degree of cluster membership information produced by FCM clustering to differentiate normal and fraudulent users. We perform an ablation study to investigate the impact of key components of the proposed method on the performance using a publicly available smart meter dataset. The results show that all key components of the proposed method contribute significantly to the performance improvement. The proposed method is compared with a set of baselines including state-of-the-art methods using smart meter data of both business users and residential users. The comparison results indicate that the proposed method achieves significantly better detection performance than all baseline methods. We also show that the proposed method maintains a good performance when the detection time frame is reduced from 30 to 20 days.
- 11 July 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 71 (00189456) , 1-10
A Novel Unsupervised Data-Driven Method for Electricity Theft Detection in AMI Using Observer Meters
The smart meter data of the advanced metering infrastructure (AMI) can be tampered by electricity thieves with advanced digital instruments or cyber attacks to reduce their electricity bills, which causes devastating financial losses to utilities. A novel unsupervised data-driven method for electricity theft detection in AMI is proposed in this article. The method incorporates observer meter data, wavelet-based feature extraction, and fuzzy c-means (FCM) clustering. A new anomaly score is developed based on the degree of cluster membership information produced by FCM clustering to differentiate normal and fraudulent users. We perform an ablation study to investigate the impact of key components of the proposed method on the performance using a publicly available smart meter dataset. The results show that all key components of the proposed method contribute significantly to the performance improvement. The proposed method is compared with a set of baselines including state-of-the-art methods using smart meter data of both business users and residential users. The comparison results indicate that the proposed method achieves significantly better detection performance than all baseline methods. We also show that the proposed method maintains a good performance when the detection time frame is reduced from 30 to 20 days.
- 1 April 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 71 (00189456) , 1-11
A New Deep Anomaly Detection-Based Method for User Authentication Using Multichannel Surface EMG Signals of Hand Gestures
User authentication plays an important role in securing systems and devices by preventing unauthorized accesses. Although surface electromyogram (sEMG) has been widely applied for human machine interface (HMI) applications, it has only seen a very limited use for user authentication. In this article, we investigate the use of multichannel sEMG signals of hand gestures for user authentication. We propose a new deep anomaly detection-based user authentication method which employs sEMG images generated from multichannel sEMG signals. The deep anomaly detection model classifies the user performing the hand gesture as client or imposter by using sEMG images as the input. Different sEMG image generation methods are studied in this article. The performance of the proposed method is evaluated with a high density sEMG (HD-sEMG) dataset and a sparse density sEMG (SD-sEMG) dataset under three authentication test scenarios. Among the sEMG image generation methods, root mean square (rms) map achieves significantly better performance than others. The proposed method with rms map also greatly outperforms the reference method, especially when using SD-sEMG signals. The results demonstrate the validity of the proposed method with rms map for user authentication.
- 1 April 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 71 (00189456) , 1-11
A New Deep Anomaly Detection-Based Method for User Authentication Using Multichannel Surface EMG Signals of Hand Gestures
User authentication plays an important role in securing systems and devices by preventing unauthorized accesses. Although surface electromyogram (sEMG) has been widely applied for human machine interface (HMI) applications, it has only seen a very limited use for user authentication. In this article, we investigate the use of multichannel sEMG signals of hand gestures for user authentication. We propose a new deep anomaly detection-based user authentication method which employs sEMG images generated from multichannel sEMG signals. The deep anomaly detection model classifies the user performing the hand gesture as client or imposter by using sEMG images as the input. Different sEMG image generation methods are studied in this article. The performance of the proposed method is evaluated with a high density sEMG (HD-sEMG) dataset and a sparse density sEMG (SD-sEMG) dataset under three authentication test scenarios. Among the sEMG image generation methods, root mean square (rms) map achieves significantly better performance than others. The proposed method with rms map also greatly outperforms the reference method, especially when using SD-sEMG signals. The results demonstrate the validity of the proposed method with rms map for user authentication.
- Open Access
- 10 September 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 9, 126065-126074
Deep Feature Fusion for Rumor Detection on Twitter
The increasing popularity of social media has made the creation and spread of rumors much easier. Widespread rumors on social media could cause devastating damages to society and individuals. Automatically detecting rumors in a timely manner is greatly needed but also very challenging technically. In this paper, we propose a new deep feature fusion method that employs the linguistic characteristics of the source tweet text and the underlying patterns of the propagation tree of the source tweet for Twitter rumor detection. Specifically, the pre-trained Transformer-based model is applied to extract context-sensitive linguistic features from the short source tweet text. A novel sequential encoding method is proposed to embed the propagation tree of a source tweet into the vector space. A convolutional neural network (CNN) architecture is then developed to extract temporal-structural features from the encoded propagation tree. The performance of the proposed deep feature fusion method is evaluated with two public Twitter rumor datasets. The results demonstrate that the proposed method achieves significantly better detection performance than other state-of-the-art baseline methods.
- 10 September 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 9, 126065-126074
Open AccessDeep Feature Fusion for Rumor Detection on Twitter
The increasing popularity of social media has made the creation and spread of rumors much easier. Widespread rumors on social media could cause devastating damages to society and individuals. Automatically detecting rumors in a timely manner is greatly needed but also very challenging technically. In this paper, we propose a new deep feature fusion method that employs the linguistic characteristics of the source tweet text and the underlying patterns of the propagation tree of the source tweet for Twitter rumor detection. Specifically, the pre-trained Transformer-based model is applied to extract context-sensitive linguistic features from the short source tweet text. A novel sequential encoding method is proposed to embed the propagation tree of a source tweet into the vector space. A convolutional neural network (CNN) architecture is then developed to extract temporal-structural features from the encoded propagation tree. The performance of the proposed deep feature fusion method is evaluated with two public Twitter rumor datasets. The results demonstrate that the proposed method achieves significantly better detection performance than other state-of-the-art baseline methods.