Tianyu Li Tianyu Li

Mr. Tianyu Li Tianyu Li


Research Focus & Metrics

Academic Career Length

3Years (2024-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
15 April 2024 in Sensors
Verified
Journal Article
Open access
PDF file
PLPeng Lei
Peng Lei
FMFanglan Ma
Fanglan Ma
CZChangsheng Zhu
Changsheng Zhu
TLTianyu Li Tianyu Li
Tianyu Li Tianyu Li
Soft sensors have been extensively utilized to approximate real-time power prediction in wind power generation, which is challenging to measure instantaneously. The short-term forecast of wind power aims at providing a reference for the dispatch of the intraday power grid. This study proposes a soft sensor model based on the Long Short-Term Memory (LSTM) network by combining data preprocessing with Variational Modal Decomposition (VMD) to improve wind power prediction accuracy. It does so by adopting the isolation forest algorithm for anomaly detection of the original wind power series and processing the missing data by multiple imputation. Based on the process data samples, VMD technology is used to achieve power data decomposition and noise reduction. The LSTM network is introduced to predict each modal component separately, and further sum reconstructs the prediction results of each component to complete the wind power prediction. From the experimental results, it can be seen that the LSTM network which uses an Adam optimizing algorithm has better convergence accuracy. The VMD method exhibited superior decomposition outcomes due to its inherent Wiener filter capabilities, which effectively mitigate noise and forestall modal aliasing. The Mean Absolute Percentage Error (MAPE) was reduced by 9.3508%, which indicates that the LSTM network combined with the VMD method has better prediction accuracy.

6 May 2025 in Measurement Science and Technology
Verified
Journal Article
Closed access
TLTianyu Li Tianyu Li
Tianyu Li Tianyu Li
CZChangsheng Zhu
Changsheng Zhu
JLJingjie Li
Jingjie Li
HCHang Cao
Hang Cao
HBHongwei Bai
Hongwei Bai
In the field of target detection, especially for UAV inspection insulator state detection, accurate identification of small targets and complex background environment is always a major challenge. Traditional methods solve this problem by strengthening feature acquisition, but ignore the increase of computational complexity and resource consumption, resulting in insufficient hardware resources of UAV. We innovatively enhance YOLOv8-N model comprehensively and propose a Feature Guided-You Only Look Once (FG-YOLO), which effectively solves the problems of background noise interference and insufficient global information capture by designing context anchor concat and C2 Locality-Aware Attention (C2fLA) modules. Secondly, we introduce lightweight neck networks such as SCDown, DySample and lightweight guided convolutional detection detector head to maintain the original performance of the model, while reducing the model parameters and computational complexity and enhancing the robustness of features. In addition, we design global channel directed attention mechanism module to improve the sensing ability of network to size targets through multi-scale feature fusion. Experiments show that the improved FG-YOLO has excellent performance and potential in real-time detection and feature capture in three public datasets, with mAP50 reaching 88.2, 99.6 and 99.8 respectively. Moreover, FG-YOLO has inference speed of 25 frames s−1 in edge device MAIX-II Axera-Pi, meeting the real-time detection requirements of insulator defects.

31 July 2025 in Measurement Science and Technology
Verified
Journal Article
Closed access
JLJingjie Li
Jingjie Li
CZChangsheng Zhu
Changsheng Zhu
TLTianyu Li Tianyu Li
Tianyu Li Tianyu Li
HCHang Cao
Hang Cao
HBHongwei Bai
Hongwei Bai
Detecting insulator defects is crucial for maintaining the reliability of power transmission and distribution systems. Addressing the challenges of high model complexity, low detection precision, and the difficulty of integrating with embedded devices, an LFCA-YOLO based on the local feature context attention enhancement mechanism is proposed, which effectively solves the problems of background noise interference and insufficient local feature capture by adding Concat_LFCA, C2f_LFCA, and PLFCA modules; the introduction of the ASCDown, CSPCLFCA, C2fCIB lightweight modules to improve the original performance of the model, while reducing the model parameters and computational complexity; Detect_SC, through the design of shared convolution, balances the contradiction between the lightweight and accuracy of the detection head. Experiments on the INSULATOR FAULTS DETECTION dataset show that the LFCA-YOLO model not only improves the AP=50 by 4.2%, but also decreases the number of model parameters and computational complexity by 17% and 12.3%, respectively, compared with YOLOv8n. Additionally, the model's ability to generalize has been confirmed through testing on both the Pascal VOC and SFID datasets. Notably, following the JOINT optimization process, the LFCA-YOLO model achieves an inference rate of 26 frames per second on the MAIX-III AXera-Pi, which satisfies the real-time detection demands for insulator defects. Our code and data are accessible via the provided GitHub.

2 July 2025 in IEEE Transactions on Instrumentation and Measurement
Verified
Journal Article
Closed access
TLTianyu Li Tianyu Li
Tianyu Li Tianyu Li
CZChangsheng Zhu
Changsheng Zhu
YWYongxin Wang
Yongxin Wang
JLJingjie Li
Jingjie Li
HCHang Cao
Hang Cao
SWSuchao Wang
Suchao Wang
Meters in electrical equipment are often potentially threatened by foreign objects suspended on power lines, which can raise the risk of short circuits or ground faults in the line, leading to problems such as loss of power supply to the meter and inaccurate readings. To address this problem, this article proposes a lightweight fast real-time detection network architecture, the lightweight multiscale faster capture detection transformer (LMFC-DETR) for detecting the presence of foreign objects in power lines, which fuses multilevel feature maps through the repeat parameter acceleration block (RPAB) and the faster gated convolution linear units (FGCLUs) modules in the backbone network, effectively reducing the computational burden and reducing the common degradation risk of deep networks. In addition, introducing the location feature interactive attention (LFIA) mechanism and multiscale integration of lightweight features (MIOLF) structure into efficient hybrid encoders, significantly enhanced the fusion and expression capabilities of Fusion-X features. Aiming at the challenges brought by low quality images and small target size datasets, this article designs Fusion-E module in MIOLF structure to retain and strengthen key information, and the global dynamic sequence fusion block (GDSFB) to facilitate deep interaction and complementarity among feature maps, thereby greatly improve the detection performance of the model. Experimental results show that LMFC-DETR achieves detection accuracies of 87.9 and 89.5 in the public and self-made foreign object detection HD dataset, respectively, with an average the frames per second (FPS) of 50, fully demonstrating its excellent performance and great potential in real-time foreign object detection tasks on power lines. Our code and the blocks we made are shown in https://github.com/Tianyu-Li-S/LMFC-DETR/tree/main/LMFC_Block

6 August 2025 in Digital Signal Processing
Verified
Journal Article
Closed access
TLTianyu Li Tianyu Li
Tianyu Li Tianyu Li
CZChangsheng Zhu
Changsheng Zhu
YWYongxin Wang
Yongxin Wang
MLMenghan Li
Menghan Li
HCHang Cao
Hang Cao
HBHongwei Bai
Hongwei Bai

23 October 2025 in Engineering Research Express
Verified
Journal Article
Closed access
HCHui Chen
Hui Chen
CZChangsheng Zhu
Changsheng Zhu
HBHongwei Bai
Hongwei Bai
TLTianyu Li Tianyu Li
Tianyu Li Tianyu Li
JLJingjie Li
Jingjie Li
ZXZhiyuan Xia
Zhiyuan Xia
Power system inspection is a key link to ensure its normal operation and stability. However, traditional detection models have weak generalization ability and usually only optimize for a single target, resulting in high detection costs and low efficiency. Models with strong generalization ability often have excessive computational overhead, making it difficult to meet the efficiency and real-time requirements of industrial applications. We propose a lightweight dynamic adjustable model LDA-YOLO based on YOLOv8-S to address the above issues. By systematically improving the model structure and feature processing mechanism, this method significantly enhances the model’s generalization and detection accuracy while maintaining its lightweight characteristics. Specifically, the improvement of the backbone is mainly achieved through the reconstruction of C2fDWCAA, which enhances the ability to capture local and global features without increasing the computational cost of the model. In addition, using GSConv and GSCDown instead of native downsampling effectively solves the significant computational overhead caused by native modules. For the Neck part of the module, the native concatenation module has poor ability to concatenate multi-scale information. We use CFCB for feature concatenation and improve the fusion effect of multi-scale features through context awareness. In addition, C2fCBB suppresses feature redundancy through low rank stage adaptive replacement. We also use DySample to replace native downsampling, enhancing the real-time monitoring efficiency of the model. Finally, in the detection head section, we use the DyHead detection head to achieve dynamic adjustability of the model. Experimental results demonstrate that LDA-YOLO achieves 31.8% and 23.2% reductions in parameters and computational complexity respectively compared to the baseline model. The model attains A P I o U = 50 scores of 76.8, 62.2, 81.4, and 75.6 on power line, insulator, foreign object, and equipment detection tasks, validating its superior performance in complex scenarios. The implementation is publicly available at https://github.com/chenhui331/LDA-YOLO .