Jie Ding
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Jie Ding has a total of 79 co-authors. Here are the top co-authors by number of shared publications:
- Min Xiao
Nanjing University of Posts and Telecommunications
9 shared publications Min Xiao
7 shared publicationsDezhi Shen
College Of Automation & College of Artificial Intelligence Nanjing University of Posts and Telecommunications Nanjing China
3 shared publications- Jing Zhao
School of Automation, Nanjing University of Posts and Telecommunications, Nanjing, Jiangsu, China
3 shared publications Min Wu
School of Automation and Artificial Intelligence, Nanjing University of Posts and Telecommunications, Nanjing, Jiangsu, China
3 shared publicationsSai Wang
School of Automation, Nanjing University of Posts and Telecommunications, Jiangsu, China
3 shared publications
Latest Publications
- 1 August 2026
- Published by AIP Publishing in Chaos: An Interdisciplinary Journal of Nonlinear Science
- Vol. 36 (8)
How higher-order interactions shape the dynamics of large-scale dual-star neural networks
Understanding large-scale neural network dynamics represents a fundamental challenge in both biological and artificial neural systems. Traditional pairwise interaction models inadequately capture the complex connectivity patterns observed in real neural networks, where groups of three or more neurons interact simultaneously. This paper presents a novel large-scale delayed neural network with dual-star architecture, incorporating both second-order and third-order interactions. Two analytical methods are employed to derive the characteristic polynomial for the proposed network: the Coates flow graph formula and the Schur complement method. Both approaches circumvent the computational intractability of conventional determinant-based methods for high-dimensional systems. Using the holistic element concept, rigorous conditions for local stability and Hopf bifurcation are established. Comprehensive numerical simulations validate theoretical findings and reveal that higher-order interactions exhibit non-monotonic effects on bifurcation thresholds, self-feedback mechanisms provide robust stabilization, and network size significantly influences bifurcation characteristics. Finally, the accurate function fitting achieved by our model provides new insights for the design and control of neural networks.
- 1 August 2026
- Published by AIP Publishing in Chaos: An Interdisciplinary Journal of Nonlinear Science
- Vol. 36 (8)
How higher-order interactions shape the dynamics of large-scale dual-star neural networks
Understanding large-scale neural network dynamics represents a fundamental challenge in both biological and artificial neural systems. Traditional pairwise interaction models inadequately capture the complex connectivity patterns observed in real neural networks, where groups of three or more neurons interact simultaneously. This paper presents a novel large-scale delayed neural network with dual-star architecture, incorporating both second-order and third-order interactions. Two analytical methods are employed to derive the characteristic polynomial for the proposed network: the Coates flow graph formula and the Schur complement method. Both approaches circumvent the computational intractability of conventional determinant-based methods for high-dimensional systems. Using the holistic element concept, rigorous conditions for local stability and Hopf bifurcation are established. Comprehensive numerical simulations validate theoretical findings and reveal that higher-order interactions exhibit non-monotonic effects on bifurcation thresholds, self-feedback mechanisms provide robust stabilization, and network size significantly influences bifurcation characteristics. Finally, the accurate function fitting achieved by our model provides new insights for the design and control of neural networks.
- 15 July 2026
- Published by Wiley in International Journal of Robust and Nonlinear Control
- Vol. 36 (16) , 7529-7544
Modeling and Prediction of Nonlinear Dynamic System Models by Deep Neural Networks and Koopman‐Based Subspace Identification
Traditional methods for modelling nonlinear dynamic systems often suffer from high computational complexity and limited prediction accuracy. To address these challenges, this paper proposes a novel Koopman operator‐based framework that integrates deep neural networks with subspace identification. The core idea of the framework is to decouple the overall modeling task by assigning distinct roles to its components: the deep neural network is dedicated to learning the nonlinear lifting map, while subspace identification handles the linear dynamic evolution in the lifted space. This approach improves computational efficiency while preserving model interpretability. Comparative experiments on both weakly nonlinear and strongly nonlinear systems show that the proposed method achieves high modeling accuracy with significantly reduced errors. Further theoretical stability analysis and validation under parameter perturbations confirm the reliability of the proposed approach, offering an effective and robust solution for nonlinear system modeling.
- 15 July 2026
- Published by Wiley in International Journal of Robust and Nonlinear Control
- Vol. 36 (16) , 7529-7544
Modeling and Prediction of Nonlinear Dynamic System Models by Deep Neural Networks and Koopman‐Based Subspace Identification
Traditional methods for modelling nonlinear dynamic systems often suffer from high computational complexity and limited prediction accuracy. To address these challenges, this paper proposes a novel Koopman operator‐based framework that integrates deep neural networks with subspace identification. The core idea of the framework is to decouple the overall modeling task by assigning distinct roles to its components: the deep neural network is dedicated to learning the nonlinear lifting map, while subspace identification handles the linear dynamic evolution in the lifted space. This approach improves computational efficiency while preserving model interpretability. Comparative experiments on both weakly nonlinear and strongly nonlinear systems show that the proposed method achieves high modeling accuracy with significantly reduced errors. Further theoretical stability analysis and validation under parameter perturbations confirm the reliability of the proposed approach, offering an effective and robust solution for nonlinear system modeling.
- 1 July 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE/CAA Journal of Automatica Sinica
- Vol. 13 (7) , 1770-1772
Random Search Deep Neural Networks Driven Koopman Subspace Modeling of Nonlinear Dynamics
- 1 July 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE/CAA Journal of Automatica Sinica
- Vol. 13 (7) , 1770-1772
Random Search Deep Neural Networks Driven Koopman Subspace Modeling of Nonlinear Dynamics
- 21 April 2026
- Published by SAGE Publications in Proceedings of the Institution of Mechanical Engineers, Part D: Journal of Automobile Engineering
NARX neural network optimized by polynomial chaos and forward orthogonal least squares for lithium-ion battery SOC estimation
Accurate estimation of the State of Charge (SOC) of lithium-ion batteries is crucial for optimizing battery performance, extending service life, and ensuring safe operation. In neural network-based SOC estimation, dynamic neural networks are indispensable, with the Nonlinear AutoRegressive with eXogenous inputs Neural Network (NARXNN) excelling in modeling lithium-ion batteries’ nonlinear time-varying dynamics. To enhance estimation performance, a dual-optimization framework is introduced. First, Forward Orthogonal Least Squares (FOLS) optimizes the NARXNN regression matrix through stepwise orthogonal regression, improving high-speed dynamic adaptability by selecting dominant time-delay terms. Subsequently, Polynomial Chaos (PC) optimizes NARXNN network weights via orthogonal polynomial decomposition, quantifying parameter uncertainties to boost steady-state precision. Under the UDDS, US06, and LA92 driving cycles, the FOLS-PC-NARXNN model achieved MAE values of 0.0255%, 0.0798%, and 0.0479%, respectively. These results represent a significant improvement over the 4.7276%, 1.6783%, and 1.8928% errors from the baseline NARXNN model and also surpass other mainstream methods in comparison. Furthermore, tests conducted at extended sampling intervals of 10, 15, and 30 s confirm the model’s strong robustness and maintained estimation accuracy under varied sampling conditions, collectively demonstrating the effectiveness of the dual-optimization strategy in enhancing SOC estimation.
- 21 April 2026
- Published by SAGE Publications in Proceedings of the Institution of Mechanical Engineers, Part D: Journal of Automobile Engineering
NARX neural network optimized by polynomial chaos and forward orthogonal least squares for lithium-ion battery SOC estimation
Accurate estimation of the State of Charge (SOC) of lithium-ion batteries is crucial for optimizing battery performance, extending service life, and ensuring safe operation. In neural network-based SOC estimation, dynamic neural networks are indispensable, with the Nonlinear AutoRegressive with eXogenous inputs Neural Network (NARXNN) excelling in modeling lithium-ion batteries’ nonlinear time-varying dynamics. To enhance estimation performance, a dual-optimization framework is introduced. First, Forward Orthogonal Least Squares (FOLS) optimizes the NARXNN regression matrix through stepwise orthogonal regression, improving high-speed dynamic adaptability by selecting dominant time-delay terms. Subsequently, Polynomial Chaos (PC) optimizes NARXNN network weights via orthogonal polynomial decomposition, quantifying parameter uncertainties to boost steady-state precision. Under the UDDS, US06, and LA92 driving cycles, the FOLS-PC-NARXNN model achieved MAE values of 0.0255%, 0.0798%, and 0.0479%, respectively. These results represent a significant improvement over the 4.7276%, 1.6783%, and 1.8928% errors from the baseline NARXNN model and also surpass other mainstream methods in comparison. Furthermore, tests conducted at extended sampling intervals of 10, 15, and 30 s confirm the model’s strong robustness and maintained estimation accuracy under varied sampling conditions, collectively demonstrating the effectiveness of the dual-optimization strategy in enhancing SOC estimation.
- 2 April 2026
- Published by Elsevier in Journal of Energy Storage
- Vol. 161
A physical-informed neural network framework for state of charge estimation of lithium battery
- 2 April 2026
- Published by Elsevier in Journal of Energy Storage
- Vol. 161
A physical-informed neural network framework for state of charge estimation of lithium battery
- 24 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Neural Networks and Learning Systems
- Vol. PP (2162237X) , 1-14
Attention-Guided Low-Light Object Detection Based on Receptive Field With Adaptive Depth Convolution
To solve the performance degradation of visible light-based object detectors affected by low-light environments, an attention-guided detection model based on adaptive depthwise convolution is proposed. First, the quality of the original image is improved, and various image information is restored by an image enhancement module named zero-reference dual-illumination deep-curve estimation++ (Zero-DiDCE++). Then, the article further introduces a self-attention mechanism and receptive field features, designing the feature self-attention (FSA) module and the sparse receptive field adaptive depth convolution (SRDConv) module. These modules work in concert to accentuate target features and improve detection accuracy. Finally, we introduce a framework named ZRF-you only look once (YOLO)v9 for object detection, which is designed to improve performance in low-light conditions. It incorporates Zero-DiDCE++ and connects it with an improved YOLOv9 network for algorithm integration. Experiments show that based on the ExDark, Dark Face, and Tiny-Person datasets, the ZRF-YOLOv9 algorithm improves the low-light detection accuracy by 1.1%, 2.0%, and 0.5% in mAP ${}_{5}{}_{0}$ . This model not only enhances the visibility of the image but also improves detection accuracy for target images in low-light scenes.
- 24 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Neural Networks and Learning Systems
- Vol. PP (2162237X) , 1-14
Attention-Guided Low-Light Object Detection Based on Receptive Field With Adaptive Depth Convolution
To solve the performance degradation of visible light-based object detectors affected by low-light environments, an attention-guided detection model based on adaptive depthwise convolution is proposed. First, the quality of the original image is improved, and various image information is restored by an image enhancement module named zero-reference dual-illumination deep-curve estimation++ (Zero-DiDCE++). Then, the article further introduces a self-attention mechanism and receptive field features, designing the feature self-attention (FSA) module and the sparse receptive field adaptive depth convolution (SRDConv) module. These modules work in concert to accentuate target features and improve detection accuracy. Finally, we introduce a framework named ZRF-you only look once (YOLO)v9 for object detection, which is designed to improve performance in low-light conditions. It incorporates Zero-DiDCE++ and connects it with an improved YOLOv9 network for algorithm integration. Experiments show that based on the ExDark, Dark Face, and Tiny-Person datasets, the ZRF-YOLOv9 algorithm improves the low-light detection accuracy by 1.1%, 2.0%, and 0.5% in mAP ${}_{5}{}_{0}$ . This model not only enhances the visibility of the image but also improves detection accuracy for target images in low-light scenes.
- 12 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 26 (4) , 5879-5890
A Fault Diagnosis Model Based on Spatial Reconstruction and Soft Pooling Local Enhancement in Small Samples
In rotating machinery fault diagnosis, the scarcity of fault samples significantly impacts the accuracy in industrial scenarios. This article proposes a novel fault diagnosis model—spatial reconstruction and soft pooling local enhancement network (SR-SPNet), specifically designed for scenarios with limited sample availability. First, the collected vibration signals are transformed into images using continuous wavelet transform (CWT), thereby enriching feature representation and enabling convolutional neural networks (CNNs) to effectively capture relevant features. Next, a multiscale dilated convolution (MDCNN) is employed to expand the receptive field. To reduce redundant computation and improve feature representation, a spatial reconstruction module (SRM) is introduced to differentiate key features. An improved residual (IR) block is then utilized to further extract crucial features, mitigating the risk of gradient vanishing during training. Additionally, a soft pooling local enhancement module (SPEM) is introduced to enhance the spatial feature representation of the feature maps. Experimental results on the CWRU public dataset and gearbox fault dataset demonstrate that, under conditions of limited sample size, SR-SPNet outperforms several comparative models, exhibiting superior diagnostic performance.
- 12 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 26 (4) , 5879-5890
A Fault Diagnosis Model Based on Spatial Reconstruction and Soft Pooling Local Enhancement in Small Samples
In rotating machinery fault diagnosis, the scarcity of fault samples significantly impacts the accuracy in industrial scenarios. This article proposes a novel fault diagnosis model—spatial reconstruction and soft pooling local enhancement network (SR-SPNet), specifically designed for scenarios with limited sample availability. First, the collected vibration signals are transformed into images using continuous wavelet transform (CWT), thereby enriching feature representation and enabling convolutional neural networks (CNNs) to effectively capture relevant features. Next, a multiscale dilated convolution (MDCNN) is employed to expand the receptive field. To reduce redundant computation and improve feature representation, a spatial reconstruction module (SRM) is introduced to differentiate key features. An improved residual (IR) block is then utilized to further extract crucial features, mitigating the risk of gradient vanishing during training. Additionally, a soft pooling local enhancement module (SPEM) is introduced to enhance the spatial feature representation of the feature maps. Experimental results on the CWRU public dataset and gearbox fault dataset demonstrate that, under conditions of limited sample size, SR-SPNet outperforms several comparative models, exhibiting superior diagnostic performance.
- 2 October 2025
- Published by Springer Nature in Ionics
- Vol. 31 (12) , 13119-13135
A dual-scale deep learning model for estimating lithium-ion battery SOC by data denoising
- 2 October 2025
- Published by Springer Nature in Ionics
- Vol. 31 (12) , 13119-13135
A dual-scale deep learning model for estimating lithium-ion battery SOC by data denoising
- 1 July 2025
- Published by Elsevier in Electric Power Systems Research
- Vol. 248
A modified ROHO-VMD and AKF-based framework for accurate harmonic detection
- 1 July 2025
- Published by Elsevier in Electric Power Systems Research
- Vol. 248
A modified ROHO-VMD and AKF-based framework for accurate harmonic detection