Haibo Yu

Haibo Yu

Auto-generated profile, based on publicly available information

Haibo Yu

Science Island Branch, Graduate School, University of Science and Technology of China, Hefei, China

Research Focus & Metrics

Academic Career Length

4Years (2023-Present)

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

Representative Publications

View All Publications
29 January 2026 in Applied Energy
Verified
Journal Article
Closed access
LCLing Chang
Ling Chang
HYHaibo Yu
Haibo Yu
MYMinghan Yang
Minghan Yang
ZZZiheng Zhang
Ziheng Zhang
SCShuai Chen
Shuai Chen
JWJianye Wang
Jianye Wang

4 April 2025 in Energy
Verified
Journal Article
Closed access
HYHaibo Yu
Haibo Yu
LCLing Chang
Ling Chang
MYMinghan Yang
Minghan Yang
SCShuai Chen
Shuai Chen
HLHuijuan Li
Huijuan Li
JWJianye Wang
Jianye Wang

28 August 2023 in IEEE Access
Verified
Journal Article
Open access
WWWeiyi Wang
Weiyi Wang
DYDongsheng Yu
Dongsheng Yu
HYHaibo Yu
Haibo Yu
MYMinghan Yang
Minghan Yang
CXChidong Xu
Chidong Xu
XFXiaodong Fang
Xiaodong Fang
Measuring the distribution of atmospheric aerosol concentration is of great significance for accurately assessing the scale of explosions and fires, the evolution of atmospheric environment, and so on. Due to the effect of atmospheric diffusion and other complex meteorological conditions, the distribution of aerosol concentration will change significantly in a short period of time, which puts higher requirements on the speed of aerosol concentration measurement. The existing measurement technology mainly uses LIDAR for intensive sampling of aerosol in a region. Due to the characteristics of rapid changes in aerosol distribution, it is difficult for this method to accurately reflect the overall aerosol distribution at a certain time. Although lidar can achieve accurate measurement of the average concentration of atmospheric aerosol more conveniently, due to longer data processing time and more measurement sampling times, the timeliness of lidar remote sensing has decreased, resulting in the problem of difficult to capture the shape of atmospheric smoke and clouds in a timely manner. Therefore, this study proposes a fast reconstruction deep network model of aerosol extinction coefficient for lidar remote sensing from the perspective of sparse sampling-reconstruction. This model eliminates the feature distribution difference between lidar return signals under sparse sampling and conventional dense sampling by using unsupervised generative adversarial networks from the perspective of transfer learning. Then, the extinction coefficient reconstruction network with the encoding-decoding structure maps the low-dimensional abstract features of aerosol concentration distribution back to the high-dimensional extinction coefficient representation space. This model greatly reduces the sampling number of lidar, thereby reducing the total time required for aerosol extinction coefficient inversion. Experimental measurements of smoke and clouds above a thermal power plant show that the proposed deep network model can reduce more than 50% of the lidar sampling times within the allowable error range. This indicates that the model has the ability to significantly improve the remote sensing measurement efficiency of lidar for atmospheric aerosols.

13 April 2026 in Annals of Nuclear Energy
Verified
Journal Article
Closed access
HYHaibo Yu
Haibo Yu
WWWeisheng Wang
Weisheng Wang
MYMinghan Yang
Minghan Yang
XSXiaolong Shen
Xiaolong Shen
PZPanpan Zhang
Panpan Zhang
JWJianye Wang
Jianye Wang

21 September 2026 in SSRN Electronic Journal
Verified
Preprint
Open access
XSXiaolong Shen
Xiaolong Shen
XSXiaofan Sun
Xiaofan Sun
HYHaibo Yu
Haibo Yu
DGDaochaun Ge
Daochaun Ge
JYJie Yu
Jie Yu
Main-pump fault diagnosis is essential for primary-loop safety in lead-bismuth-cooled reactors. However, the shortage of labeled in-service fault data and differences between simulated or rig-measured fault signals and real signals together hinder model transfer and reduce diagnostic performance. To address these challenges, this paper proposes Physics-Guided Health-Reference Fault Diagnosis (PGHR-FD) framework, which uses no real fault samples during model development. The core idea of PGHR-FD is to use real healthy references to guide physics-based fault synthesis. Specifically, healthy records provide normalization scale, resonance bands, and condition-matched backgrounds. Randomized physics-based fault residuals are added to these backgrounds to generate labeled pseudo-fault samples for training a network combining health-referenced envelope-order features and characteristic-frequency evidence. During diagnosis, network outputs for anomalous samples identified by a health gate are averaged across four sub-windows. Predictions are revised only when physical evidence supports the same fault class in at least three sub-windows. CWRU and Paderborn datasets serve as surrogate benchmarks for evaluating the proposed method. PGHR-FD achieves mean accuracies of 95.01 ± 1.98% across eight CWRU tasks and 99.21 ± 0.79% on the Paderborn. The results demonstrate the feasibility and effectiveness of the proposed method.