Wenyuan Wang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Wenyuan Wang has a total of 54 co-authors. Here are the top co-authors by number of shared publications:
- Wenxu Yan
Jiangnan University
23 shared publications - Hongyu Ni
State Grid Shaoxing Power Supply Company, Shaoxing, China
5 shared publications - Kutluyil Dogancay
STEM University of South Australia Mawson Lakes Boulevard 5095 Adelaide Australia
5 shared publications - Dawei Liu
School of Internet of Things Engineering, Jiangnan University, Wuxi, China
3 shared publications Dongcheng Zhou
Shaoxing Daming Electric Power Construction Company, Ltd., Shaoxing, China
2 shared publicationsShaodian Liu
Jiangnan University
2 shared publications
Latest Publications
- Open Access
- 11 July 2026
- Published by MDPI AG in World Electric Vehicle Journal
- Vol. 17 (7) , 357
Real-Time On-MCU Open-Circuit Fault Diagnosis of Electric-Vehicle Inverters Using a Lightweight Angular Sector-Energy Network
Power-switch open-circuit (OC) faults distort electric-vehicle (EV) inverter phase currents and require fast on-board diagnosis for fault-tolerant control. Trajectory-image methods encode the – current-vector trajectory as a binary image and classify it with a convolutional neural network (CNN); however, the baseline uses parameters and multiply–accumulate (MAC) operations per inference, which is costly for motor-control microcontrollers (MCUs). Here, each one-cycle trajectory is represented by a 36-dimensional normalized angular sector-energy vector and classified by a compact two-stage multilayer perceptron. Sector accumulation averages zero-mean measurement noise in the representation, without relying on noise-augmented training. The locating stage uses parameters and MACs per inference, and fewer than the baseline CNN; the complete pipeline runs on a TI F28379D in ms. On measured resistive-load currents, both methods reach accuracy from 40 to 20 dB, whereas the proposed method remains more accurate at 15 and 10 dB, including under phase-current unbalance. A supplementary balanced -load experiment preserves clean accuracy, confirming MCU-executable diagnosis under a lagging power-factor load for embedded EV inverter protection.- 11 July 2026
- Published by MDPI AG in World Electric Vehicle Journal
- Vol. 17 (7) , 357
Open AccessReal-Time On-MCU Open-Circuit Fault Diagnosis of Electric-Vehicle Inverters Using a Lightweight Angular Sector-Energy Network
Power-switch open-circuit (OC) faults distort electric-vehicle (EV) inverter phase currents and require fast on-board diagnosis for fault-tolerant control. Trajectory-image methods encode the – current-vector trajectory as a binary image and classify it with a convolutional neural network (CNN); however, the baseline uses parameters and multiply–accumulate (MAC) operations per inference, which is costly for motor-control microcontrollers (MCUs). Here, each one-cycle trajectory is represented by a 36-dimensional normalized angular sector-energy vector and classified by a compact two-stage multilayer perceptron. Sector accumulation averages zero-mean measurement noise in the representation, without relying on noise-augmented training. The locating stage uses parameters and MACs per inference, and fewer than the baseline CNN; the complete pipeline runs on a TI F28379D in ms. On measured resistive-load currents, both methods reach accuracy from 40 to 20 dB, whereas the proposed method remains more accurate at 15 and 10 dB, including under phase-current unbalance. A supplementary balanced -load experiment preserves clean accuracy, confirming MCU-executable diagnosis under a lagging power-factor load for embedded EV inverter protection. - 18 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 75 (00189456) , 9004414
Hybrid Supervised Physics-Informed Neural Network for Robust Power System State Estimation
Power System State Estimation (PSSE) is fundamental for the secure operation of power grids, relying on measurements to estimate real-time conditions. However, conventional Weighted Least Squares (WLS) is vulnerable to non-Gaussian noise and bad data. Meanwhile, existing data-driven and Physics-Guided Deep Learning (PGDL) methods often suffer from poor interpretability, reliance on ”soft” constraints, and systematic biases from simplified models. To address these limitations, this paper proposes a Hybrid Supervised Physics-Informed Neural Network (HSPINN). The architecture integrates a differentiable power flow model as a ”hard” constraint within the decoder to ensure physical consistency. A hybrid supervision mechanism, combining physics-consistency and true-state supervision losses, mitigates model-induced biases, while establishing a theoretical analysis of the framework’s structural robustness against stealthy False Data Injection Attacks (FDIAs), which is further hardened via adversarial training. Furthermore, an uncertainty-based adaptive weighting scheme dynamically balances these competing losses. To evaluate the performance of the proposed framework, a comparison is made between the traditional Mean Squared Error (MSE) loss and robust alternatives, including the Huber and Generalized Correntropy Loss (GCL) functions. Simulation results on the IEEE 14-bus and 118-bus systems, tested under various scenarios with Gaussian and non-Gaussian noise, severe bad data, and stealthy FDIA, show that the proposed HSPINN framework significantly outperforms the traditional WLS method in both accuracy and robustness.
- 18 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 75 (00189456) , 9004414
Hybrid Supervised Physics-Informed Neural Network for Robust Power System State Estimation
Power System State Estimation (PSSE) is fundamental for the secure operation of power grids, relying on measurements to estimate real-time conditions. However, conventional Weighted Least Squares (WLS) is vulnerable to non-Gaussian noise and bad data. Meanwhile, existing data-driven and Physics-Guided Deep Learning (PGDL) methods often suffer from poor interpretability, reliance on ”soft” constraints, and systematic biases from simplified models. To address these limitations, this paper proposes a Hybrid Supervised Physics-Informed Neural Network (HSPINN). The architecture integrates a differentiable power flow model as a ”hard” constraint within the decoder to ensure physical consistency. A hybrid supervision mechanism, combining physics-consistency and true-state supervision losses, mitigates model-induced biases, while establishing a theoretical analysis of the framework’s structural robustness against stealthy False Data Injection Attacks (FDIAs), which is further hardened via adversarial training. Furthermore, an uncertainty-based adaptive weighting scheme dynamically balances these competing losses. To evaluate the performance of the proposed framework, a comparison is made between the traditional Mean Squared Error (MSE) loss and robust alternatives, including the Huber and Generalized Correntropy Loss (GCL) functions. Simulation results on the IEEE 14-bus and 118-bus systems, tested under various scenarios with Gaussian and non-Gaussian noise, severe bad data, and stealthy FDIA, show that the proposed HSPINN framework significantly outperforms the traditional WLS method in both accuracy and robustness.
- 18 March 2026
- Published by Elsevier in Expert Systems with Applications
- Vol. 319
SDGNN: Enhancing simplified deep graph neural networks with adaptive aggregation combination mechanism
- 18 March 2026
- Published by Elsevier in Expert Systems with Applications
- Vol. 319
SDGNN: Enhancing simplified deep graph neural networks with adaptive aggregation combination mechanism
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Signal Processing Letters
- Vol. 33, 3956-3960
Bounded Planes and Wall Corners With Hierarchical Association for LiDAR-Inertial Odometry
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Signal Processing Letters
- Vol. 33, 3956-3960
Bounded Planes and Wall Corners With Hierarchical Association for LiDAR-Inertial Odometry
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Signal Processing Letters
- Vol. 33, 3706-3710
Node-Wise Adaptive Scale Fusion: A Dual-View Framework for Spatio-Temporal Graphs
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Signal Processing Letters
- Vol. 33, 3706-3710
Node-Wise Adaptive Scale Fusion: A Dual-View Framework for Spatio-Temporal Graphs
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Signal Processing Letters
- Vol. 33, 2874-2878
Wall-Object Multimodal Encoding for Robust Coarse-to-Fine Relocalization Using LiDAR Signals
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Signal Processing Letters
- Vol. 33, 2874-2878
Wall-Object Multimodal Encoding for Robust Coarse-to-Fine Relocalization Using LiDAR Signals
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Signal Processing Letters
- Vol. 33 (10709908) , 1646-1650
ARMA-Based Graph Filtering for Robust Spatio-Temporal Prediction in MTGNNs
The MixHop convolution employed in multivariate time series graph neural networks (MTGNNs) predominantly relies on polynomial-based graph filters, which constrain spatial receptive fields, limit spectral expressivity, and diminish robustness to graph noise or non-stationarity. To overcome these limitations, ARMA-MTGNN is proposed, in which the MixHop module is replaced by a graph convolutional skip (GCS) layer, an approximation of autoregressive–moving-average (ARMA) graph filtering. This design facilitates global spatial aggregation while retaining ARMA's advantages in frequency-domain modelling and stability. Theoretical analysis demonstrates that ARMA graph convolution provides larger receptive fields, richer spectral representations, and enhanced robustness to structural perturbations compared with MixHop. Experimental evaluation on four benchmark datasets indicates that ARMA-MTGNN consistently outperforms MTGNN and achieves competitive or superior performance relative to state-of-the-art baselines, offering a parameter-efficient, expressive, and robust framework for multivariate time series forecasting.
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Signal Processing Letters
- Vol. 33 (10709908) , 1646-1650
ARMA-Based Graph Filtering for Robust Spatio-Temporal Prediction in MTGNNs
The MixHop convolution employed in multivariate time series graph neural networks (MTGNNs) predominantly relies on polynomial-based graph filters, which constrain spatial receptive fields, limit spectral expressivity, and diminish robustness to graph noise or non-stationarity. To overcome these limitations, ARMA-MTGNN is proposed, in which the MixHop module is replaced by a graph convolutional skip (GCS) layer, an approximation of autoregressive–moving-average (ARMA) graph filtering. This design facilitates global spatial aggregation while retaining ARMA's advantages in frequency-domain modelling and stability. Theoretical analysis demonstrates that ARMA graph convolution provides larger receptive fields, richer spectral representations, and enhanced robustness to structural perturbations compared with MixHop. Experimental evaluation on four benchmark datasets indicates that ARMA-MTGNN consistently outperforms MTGNN and achieves competitive or superior performance relative to state-of-the-art baselines, offering a parameter-efficient, expressive, and robust framework for multivariate time series forecasting.