Robust Clustering and Anomaly Detection of User Electricity Consumption Behavior Based on Correntropy

IET Generation Transmission & Distribution, Год журнала: 2025, Номер 19(1)

Опубликована: Янв. 1, 2025

ABSTRACT Anomaly detection in power systems is crucial for ensuring the safety and stability of electrical grids. Traditional methods struggle to extract meaningful features from electricity consumption data due significant differences usage patterns across various user types, such as residential industrial users. Applying a single model all categories increases feature complexity computational demands. Additionally, non‐Gaussian outliers caused by equipment measurement noise can significantly deviate normal patterns, making them difficult filter using standard methods. To address these challenges, this paper proposes robust, user‐type‐specific anomaly method. After preprocessing, correntropy‐based K‐means clustering method used separate users with noisy data. A two‐stage framework combining fuzzy logic convolutional neural network (CNN)‐long short‐term memory (LSTM) enhances both efficiency accuracy. The experiments were conducted open‐source datasets, results demonstrated that our achieved an accuracy 95%, which approximately 4% higher than traditional Isolation Forest This indicates approach effectively balances detection, its generalizability further validated on additional dataset.

Язык: Английский

---

Multi‐Channel Deep Pulse‐Coupled Net: A Novel Bearing Fault Diagnosis Framework

IET Image Processing, Год журнала: 2025, Номер 19(1)

Опубликована: Янв. 1, 2025

ABSTRACT Bearings are a critical part of various industrial equipment. Existing bearing fault detection methods face challenges such as complicated data preprocessing, difficulty in analysing time series data, and inability to learn multi‐dimensional features, resulting insufficient accuracy. To address these issues, this study proposes novel diagnosis model called multi‐channel deep pulse‐coupled net (MC‐DPCN) inspired by the mechanisms image processing primary visual cortex brain. Initially, transformed into greyscale spectrograms, allowing handle effectively. The method introduces convolutional coupling mechanism between multiple channels, enabling framework can features on all channels well. This conducted experiments using dataset from Case Western Reserve University. On dataset, 6‐channel (adjustable specific tasks) MC‐DPCN was utilized analyse one normal class three classes. Compared state‐of‐the‐art methods, our demonstrates highest diagnostic accuracies. achieved an accuracy 99.96% vs. discrimination 99.89% type (average result ten‐fold cross‐validation).

Язык: Английский

---

Towards trustworthy civil aviation hazards identification: An uncertainty-aware deep learning framework

Advanced Engineering Informatics, Год журнала: 2025, Номер 65, С. 103280 - 103280

Опубликована: Март 29, 2025

Язык: Английский

---

Nuclear Power Systems Unsupervised Anomaly Localization Considering Spatiotemporal Information and Influence Mechanism between Devices

Energy, Год журнала: 2025, Номер unknown, С. 136204 - 136204

Опубликована: Апрель 1, 2025

Язык: Английский

---

Stochastic Embedding Domain Generalization Network for Rotating Machinery Fault Diagnosis under Unseen Operating Conditions

IEEE Sensors Journal, Год журнала: 2024, Номер 24(11), С. 17846 - 17855

Опубликована: Апрель 9, 2024

Domain generalization-based fault diagnosis methods have been extensively explored in cross-domain under various operating conditions recent times. Nevertheless, these adhere to a common premise that the modes across each available source domain remain consistent. The label inconsistent problem arises when model extracts domain-invariant features from multiple domains. That is say, between domains are inconsistent, resulting overfitting scarce during training. Aiming at this problem, study presents Stochastic Embedding Generalization Network (SEDGN) for rotating machinery diagnosis, particularly scenarios where exist Firstly, stochastic embedding layer designed mitigate modes, which weights of identifier mode modeled by Gaussian distributions and will be optimized Secondly, ground-truth label-guided correlation alignment further introduced shared domains, enhancing feature extraction modes. Finally, gearbox dataset containing bearing gear faults utilized simulate generalization tasks problems, effectiveness proposed SEDGN validated.

Язык: Английский

---

Deep learning with local spatiotemporal structure preserving for soft sensor development of complex industrial processes

Applied Soft Computing, Год журнала: 2024, Номер 164, С. 111974 - 111974

Опубликована: Июль 6, 2024

Язык: Английский

---

Adaptive Attention-Driven Few-Shot Learning for Robust Fault Diagnosis

IEEE Sensors Journal, Год журнала: 2024, Номер 24(16), С. 26034 - 26043

Опубликована: Авг. 15, 2024

Язык: Английский

---

Intelligent multi-severity nuclear accident identification under transferable operation conditions

Annals of Nuclear Energy, Год журнала: 2024, Номер 201, С. 110416 - 110416

Опубликована: Фев. 15, 2024

Язык: Английский

---

Online Knowledge Distillation for Machine Health Prognosis Considering Edge Deployment

IEEE Internet of Things Journal, Год журнала: 2024, Номер 11(16), С. 27828 - 27839

Опубликована: Май 22, 2024

Complex neural networks with deep structures are beneficial for solving problems such as fault classification and health prediction of industrial equipment due to their powerful feature extraction capabilities. Unfortunately, corresponding complex models designed based on learning algorithms require huge computational memory resources, making them difficult achieve effective edge deployment. In order solve this difficulty practical significance, paper proposes an online knowledge distillation framework machine prognosis. Within framework, the learned can be distilled simple that deployed devices in sites. Specifically, response-based module, feature-based relation-based module respectively information transmission from different levels. Furthermore, inherent differences between have been fully considered impact efficiency distillation, adaptive mutual strategy has contrapuntally proposed address limitation. Multiple experiments were conducted two sets run-to-failure datasets mechanical key components pairs verify effectiveness framework. The experimental results show student-networks effectively improve performance after receiving teacher-networks, providing a new solution prognosis under premise

Язык: Английский

---

Enhancing resilience in complex energy systems through real-time anomaly detection: a systematic literature review

Energy Informatics, Год журнала: 2024, Номер 7(1)

Опубликована: Окт. 4, 2024

Язык: Английский

---