Aug 2026· Engineering Research Express· Vol 8, pp. 165328· 0 citations· 25 references
Physics
TL;DR
A deep learning framework based on a Mamba-driven state-space model architecture for comprehensive PQ disturbance classification is proposed and results indicate that the proposed method is well-suited for real-time smart grid monitoring and intelligent protection systems.
Abstract
The modernization of electrical power systems has accelerated the transition from conventional centralized grids to intelligent, cyber-enabled smart grids characterized by bidirectional energy flow, advanced sensing, and real-time data-driven control. However, the large-scale integration of inverter-interfaced renewable energy sources and nonlinear loads has introduced significant power quality (PQ) challenges, including voltage sag, swell, harmonics, flicker, transients, and composite disturbances under noisy conditions. Accurate and computationally efficient detection of such disturbances is critical for intelligent digital relaying and real-time monitoring applications. To address these challenges, this paper proposes a deep learning framework based on a Mamba-driven state-space model architecture for comprehensive PQ disturbance classification. The proposed approach converts voltage waveforms into structured two-dimensional patches using convolutional operations, which are then transformed into sequential representations for efficient long-sequence modeling via the Mamba module. Spatial feature extraction and temporal dependency learning are jointly achieved, while maintaining linear-time complexity and reduced memory requirements compared to transformer-based architectures. The extracted latent features are subsequently classified using a deep neural network (DNN)followed by softmax classifier. Eighteen PQ disturbance categories, defined in accordance with IEEE standards, are considered, including single and composite disturbances under noisy operating conditions. Experimental evaluation examines feature representations, convergence characteristics, confusion matrix performance, computational time, and memory usage. Comparative results demonstrate that the proposed Mamba-based framework achieves high classification accuracy with significantly lower computational complexity than conventional transformer models. The findings indicate that the proposed method is well-suited for real-time smart grid monitoring and intelligent protection systems.
: The accelerating penetration of variable renewable generation into electrical grids necessitates advanced forecasting paradigms capable of addressing stochastic intermittency challenges. This paper engineers a novel hybrid intelligent framework—integrating Improved Lotus Effect Algorithm (ILEA), Variational Mode Decomposition (VMD), and ensemble deep learning—specifically designed for ultra-short-term wind power prediction in energy dispatch applications. Unlike conventional approaches relying on static parameterization, the proposed methodology employs elite chaotic opposition-based learning to autonomously optimize VMD decomposition levels, thereby decoupling non-stationary wind power sequences into stationary sub-components without empirical intervention. Local temporal feature extraction is subsequently performed via one-dimensional Convolutional Neural Networks (1D-CNN), wherein sliding convolutional kernels operate along the temporal axis to capture localized patterns—such as ramp events, gradient transitions, and short-term fluctuations—embedded within the univariate wind power sequence. Temporal dependencies are subsequently modeled through Bidirectional Long Short-Term Memory (BiLSTM) networks enhanced with attention mechanisms. An Adaptive Boosting (AdaBoost) ensemble strategy further aggregates multiple weak regressors to fortify prediction robustness against ramp events and turbulent meteorological conditions. Comprehensive validation utilizing Belgian grid operational data demonstrates that the proposed architecture achieves substantial error reductions, attaining Root Mean Square Error (RMSE) of 2.2496 MW, Mean Absolute Error (MAE) of 1.9897 MW, and Symmetric Mean Absolute Percentage Error (SMAPE) of 5.26%, with a Coefficient of Determination (R) 2 coefficient approaching unity (0.9962). These results underscore the framework’s practical efficacy for grid stabilization, load frequency control, and energy management decision support systems operating under high renewable penetration scenarios.
Lei Shen, Qifeng Xiang, Q. Gao et al.· Energy Engineering· 0 citations
With the continuous penetration of renewable energy and power electronic equipment into modern power systems, the occurrence frequency of composite power quality disturbances has increased significantly. The accurate classification of various composite disturbances under strong noise remains a critical technical challenge. The existing single time–frequency transformation methods cannot simultaneously capture transient time-domain details and fine frequency-domain features of steady-state harmonics, while mainstream deep learning classification networks contain redundant parameters and introduce excessive computational overhead, failing to meet the real-time deployment requirements of power edge terminals. To address these limitations, a lightweight Coordinate Attention ResNet network named ResNet–LCA is proposed based on the dual-modal time–frequency fusion of the Continuous Wavelet Transform and Short-Time Fourier Transform. First, the two transforms are implemented separately to generate two groups of complementary time–frequency maps, which are concatenated along the channel dimension to fully extract the coupling features between the steady-state harmonics and the transient impulses. Second, a Haar wavelet subband mean aggregation module is designed for dimensionality reduction with negligible information loss. This module eliminates the channel redundancy introduced by the multimodal fusion and reduces the overall computational overhead at the input stage. Finally, a lightweight residual network integrated with Coordinate Attention is constructed, with Grouped Half-Convolution adopted to compress the model parameters. CA offsets the feature attenuation induced by the lightweight structural design and further improves the model’s noise immunity. A simulation verification was carried out on a simulated dataset covering 25 types of single and superimposed composite disturbances. At a signal-to-noise ratio of 20 dB, the proposed method achieved an average classification accuracy of 97.92%, with only 5.32 M total parameters and a single-sample GPU inference latency of 0.33 ms. Compared with standard ResNet-18 under 20 dB noisy conditions, the total parameter volume was reduced by 52.7%, the inference latency was shortened by 0.13 ms, and the classification accuracy was improved by 0.60 percentage points. The proposed method achieves coordinated optimization of classification accuracy, noise immunity and inference efficiency, and it can provide lightweight technical support for online intelligent power quality monitoring at the edge nodes of microgrids and islanded power systems.
Electric energy theft is a major issue for the sustained use of modern smart grids (SGs). It affects the electric system’s overall long-term reliability and affordability. Even though advanced metering infrastructure can gather a large amount of data regarding electric energy utilization, it is quite difficult to address non-technical losses (NTLs). This is because electric energy consumption patterns exhibit high-dimensional characteristics, time-varying features, and are not invariably steady. Adapting the simple machine learning and artificial intelligence approach generally requires a single multidimensional electric energy consumption estimation and topical feature extraction mechanism, a limitation that prevents it from characterizing long-range temporal dependencies and intermittent behavior. Consequently, such procedures tend to have lower detection and elevated rates of false positives. In this regard, in this research, a novel transformer-based wide and deep convolutional neural network (TWiDeCNN) is proposed to efficiently identify electric energy theft in a scenario based on SGs. The suggested TWiDeCNN model is trained and tested on a real-world dataset of electric energy use collected from working SG environments, which makes it useful in the real world. Experimental results show that the suggested model works better than modern advanced methods that use binary classification metrics. These results show that the model performs well, is stable, and can be used at a large scale. They also show that it could be used in cutting-edge ways to improve energy management and find more electric energy theft. In-depth simulation results on the State Grid Corporation of China (SGCC) dataset demonstrate that TWiDeCNN outperforms the original wide model. In addition, it outperforms deep convolutional neural network (CNN) and other benchmarks in terms of ranking capability and detection performance with MAP@100 of
96.00
%
and MAP@200 of
93.58
%
at a 70% split. Moreover, parameter sensitivity analysis shows that proper tuning of parameters
α
,
β
,
γ
, and
R
would help the model build a deep feature-based, powerful representation and a better adaptation ability. Our proposed model maintains robust performance, demonstrating its stability and effectiveness for electric energy theft detection.
Hassan Ali Khan, Zahid Wadud, Ghulam Hafeez et al.· Frontiers in Energy Efficien...· 0 citations
: Sub-synchronous oscillation (SSO) is a critical stability threat in wind farms connected to power grids through series-compensated transmission lines, where delayed or inaccurate recognition may lead to converter overcurrent, turbine disconnection, shaft torsional vibration, and large-scale power fluctuations. Existing model-based approaches depend heavily on complete system parameters, while conventional signal-processing methods often require long observation windows and are less suitable for rapid online warning. To address these limitations, this paper proposes a physics-aware data-to-image recognition framework and a transfer-learning-based prior VGG model for identifying SSO hazard levels in series-compensated doubly fed induction generator wind farms. First, multi-source power-quality variables, including voltage, current, frequency, active and reactive power, voltage total harmonic distortion, current total harmonic distortion, and rotor-speed-related information, are organized into a two-dimensional feature matrix according to the actual topology of the DFIG grid-connected system. This matrix preserves the electrical relationships defined by the DFIG grid structure and effectively enhances discriminative feature representation by aligning heterogeneous measurements with grid-side and wind-farm-cluster connectivity, enabling more structured feature extraction for subsequent learning models. The resulting matrix is then normalized and mapped into RGB images, allowing neural networks to learn both local physical consistency and cross-cluster spatial coupling patterns. Second, feedforward neural network, capsule neural network, and VGG-style convolutional models are developed and compared for three SSO states: attenuating oscillation, constant-amplitude oscillation, and diverging oscillation. To alleviate limited SSO samples and deep network convergence difficulty, simplified prior SSO images are generated from physically interpretable class prototypes and used to pre-train the VGG feature extractor before transfer to simulated SSO images. The proposed framework achieves strong recognition performance under the random sample-level protocol and shows the best independent case-level performance among the compared methods, providing a useful reference for SSO analysis of wind turbines.
Xinmeng Zhou, Jing Shi, Zhenping Yu et al.· Energy Engineering· 0 citations
ABSTRACT -The increasing integration of renewable energy resources, distributed generation, electric vehicles, and intelligent monitoring devices has significantly enhanced the complexity of modern smart grids, making conventional fault detection techniques inadequate for ensuring reliable and secure power system operation. This paper presents an AI-based fault detection framework for smart grids that utilizes machine learning and deep learning techniques to identify, classify, and localize electrical faults in real time. The proposed framework collects operational data from intelligent electronic devices (IEDs), phasor measurement units (PMUs), smart meters, and IoT-enabled sensors. The acquired data undergo preprocessing, normalization, and feature extraction before being analyzed using a hybrid deep learning model combining Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks. The CNN effectively extracts spatial fault characteristics, while the LSTM captures temporal variations in electrical signals for accurate fault prediction. The trained model is deployed on an edge-cloud architecture to enable low-latency fault detection, rapid decision-making, and remote monitoring. Experimental evaluation demonstrates that the proposed system achieves high fault detection accuracy, reduced false alarm rates, and faster response times compared with conventional rule-based and statistical approaches. The framework also enhances grid reliability, minimizes outage duration, supports predictive maintenance, and improves operational efficiency under dynamic grid conditions. These results indicate that artificial intelligence can significantly strengthen the resilience and automation of next-generation smart grid infrastructures.
Keywords— Smart Grid, Artificial Intelligence (AI), Fault Detection, Deep Learning, Machine Learning, Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), Internet of Things (IoT), Predictive Maintenance, Edge Computing.
T. Anvesh, Akshaya Chelpuri, Ambati Chandu· International Scientific Jou...· 0 citations
The stable detection of faults in smart power grids is essential when focusing on the stable operation and the reduction of the downtime. In this paper, the author suggests a hybrid deep learning-based model that combines convolutional neural networks (CNN) and long short-term memory (LSTM) with explainable artificial intelligence (XAI) to detect and classify faults accurately and interpretably. The model aims at capturing the spatial and time-varying attributes of multivariate electrical signatures such as voltage, current, and frequency changes. A combination of real time sensor measurements and simulated fault conditions are used which includes several fault classes including LG, LL, LLG, and three phase faults. Experimental evaluation demonstrates that the proposed model achieves a classification accuracy of 97.84%, with an F1-score of 97.08%, outperforming conventional CNN and LSTM models by more than 3%. The framework is also able to work in a noisy environment with a stable performance of less than 2% performance decrease, and inference latency of 18 ms, which can be deployed in real-time. Besides, the combination of SHAP and attention processes improves the interpretability through the detection of the crucial features that lead to fault prediction. The findings suggest that the suggested solution is a powerful, scalable, and clear solution when it comes to intelligent fault management in a contemporary smart grid.
Udit Mamodiya, Divyanshu Sinha, I. Kishor et al.· Scientific Reports· 0 citations