Xu Rui

Dr. Xu Rui

Shenzhen International Graduate School, Tsinghua University, Shenzhen, Guangdong, China
orcid

Research Focus & Metrics

Academic Career Length

2Years (2025-Present)

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

Representative Publications

View All Publications
21 July 2025 in Symmetry
Verified
Journal Article
Open access
PDF file
ZHZhikai He
Zhikai He
Xu RuiXu Rui
Xu Rui
BWBinyu Wang
Binyu Wang
QMQisong 无 Meng
Qisong 无 Meng
QTQiang Tang
Qiang Tang
DJDuan Jianyu
Duan Jianyu
The Internet of Vehicles (IoV) operates in an environment characterized by asymmetric security threats, where centralized vulnerabilities create a critical imbalance that can be disproportionately exploited by attackers. This study addresses this imbalance by proposing a symmetrical security framework that integrates Blockchain and Federated Learning (FL) to restore equilibrium in the Vehicle–Road–Cloud ecosystem. The evolution toward sixth-generation (6G) technologies amplifies both the potential of vehicle-to-everything (V2X) communications and its inherent security risks. The proposed framework achieves a delicate balance between robust security and operational efficiency. By leveraging blockchain’s symmetrical and decentralized distribution of trust, the framework ensures data and model integrity. Concurrently, the privacy-preserving approach of FL balances the need for collaborative intelligence with the imperative of safeguarding sensitive vehicle data. A novel Cloud Proxy Re-Encryption Offloading (CPRE-IoV) algorithm is introduced to facilitate efficient model updates. The architecture employs a partitioned blockchain and a smart contract-driven FL pipeline to symmetrically neutralize threats from malicious nodes. Finally, extensive simulations validate the framework’s effectiveness in establishing a resilient and symmetrically secure foundation for next-generation IoV networks.
1

23 December 2025 in Journal of King Saud University Computer and Information Sciences
Verified
Journal Article
Open access
PDF file
RBRui Bai
Rui Bai
Xu RuiXu Rui
Xu Rui
TRTeng Rui
Teng Rui
JLJiale Liu
Jiale Liu
Hoi Leong LeeHoi Leong Lee
Hoi Leong Lee
FYFujiang Yuan
Fujiang Yuan
Autonomous driving technology has made significant advancements in recent years, yet challenges remain in ensuring safe and comfortable interactions with human-driven vehicles (HDVs), particularly during lane-changing maneuvers. This paper proposes an improved double quintic polynomial approach for safe and efficient lane-changing in mixed traffic environments. The proposed method integrates a time-to-collision (TTC) based evaluation mechanism directly into the trajectory optimization process, ensuring that the ego vehicle proactively maintains a safe gap from surrounding HDVs throughout the maneuver. The framework comprises state estimation for both the autonomous vehicle (AV) and HDVs, trajectory generation using double quintic polynomials, real-time TTC computation, and adaptive trajectory evaluation. To the best of our knowledge, this is the first work to embed an analytic TTC penalty directly into the closed-form double-quintic polynomial solver, enabling real-time safety-aware trajectory generation without post-hoc validation. Extensive simulations conducted under diverse traffic scenarios demonstrate the safety, efficiency, and comfort of the proposed approach compared to conventional methods such as quintic polynomials, Bezier curves, and B-splines. The results highlight that the improved method not only avoids collisions but also ensures smooth transitions and adaptive decision-making in dynamic environments. This work bridges the gap between model-based and adaptive trajectory planning approaches, offering a stable solution for real-world autonomous driving applications.

27 April 2026 in Scientific Reports
Verified
Journal Article
Open access
Xu RuiXu Rui
Xu Rui
YZYu Zheng
Yu Zheng
QMQisong Meng
Qisong Meng
QYQiyu Yang
Qiyu Yang
As renewable energy penetration in power grids increases, inherent intermittency and volatility pose severe stability challenges. Traditional assessment methods face computational complexity bottlenecks, while machine learning (ML) models, despite strong predictive performance, suffer from “black-box” opacity limiting adoption in safety-critical systems. This study proposes a comprehensive ML framework integrating LIME interpretability and physical mechanism validation for renewable energy grid stability assessment, using a publicly available synthetic benchmark dataset (2000 samples, 15 features) adhering to IEC/IEEE standards. We systematically compare ten classification models spanning traditional ML (logistic regression, K-nearest neighbors, support vector machine, random forest, gradient boosting) and deep learning (MLP, 1D CNN, LSTM, GRU, tabular ResNet). Results demonstrate traditional methods’ competitive performance on this small-scale tabular dataset—gradient boosting achieves 84.5% accuracy, ROC AUC 0.904, MCC 0.626, outperforming deep architectures constrained by overfitting on 1600 training samples. We emphasize that these findings are specific to the limited dataset scale (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} $$n=1600$$); deep learning may demonstrate superior performance with substantially larger training sets (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} $$n > 10,000$$). We develop a four-level LIME interpretability framework: (i) single-sample local interpretation, (ii) global feature importance aggregation identifying FreqDeviation and HarmonicTHD as most critical factors (contributions > 10), (iii) contribution directionality analysis, and (iv) multi-sample consistency verification (CV < 15%). Rationale for CV < 15%: This threshold balances coverage (88% of samples) with reliability, filtering unstable interpretations while retaining robust feature rankings, validated through empirical testing and expert assessment. LIME-identified features align with power system theory—frequency deviation reflects active power imbalance, harmonic distortion affects damping. However, we identify spurious correlations (Sample 127: low harmonics paradoxically promoting instability), underscoring expert validation necessity before deployment. Exploratory LLM integration (GPT-4) demonstrates feasibility of converting LIME outputs into actionable natural language recommendations, achieving expert-rated technical accuracy 4.2 ± 0.6/5. We explicitly acknowledge critical limitations: (i) synthetic data may not capture all real-world complexities (rare failure modes, cascading failures, stochastic weather patterns), requiring validation on actual SCADA/PMU data; (ii) the evaluated deep learning architectures exclude recent state-of-the-art tabular models (e.g., FT-transformer, TabNet), limiting the scope of our deep learning assessment; (iii) model performance may degrade by 5–15% on real data without domain adaptation. This framework advances explainable AI for power systems, providing pragmatic guidance for practitioners and establishing rigorous interpretability methodology transferable to physics-constrained ML tasks.

29 October 2026 in Applied Sciences
Verified
Journal Article
Open access
PDF file
FZFengwei Zhang
Fengwei Zhang
Xu RuiXu Rui
Xu Rui
QTQiang Tang
Qiang Tang
Contact-rich precision assembly fails in ways visible in the force/torque and proprioceptive streams before they are visible in the finished part, which makes online quality monitoring attractive. We ask whether monitoring accuracy or the recovery policy is the binding constraint on closed-loop performance, and answer it by improving both. For monitoring we propose AdaStack, which mixes a random forest on summary statistics, a fully convolutional network and a CNN-LSTM with simplex weights fitted on pooled out-of-fold posteriors, so the combination adapts to whichever learner suits a given task. On the UCI Robot Execution Failures benchmark, evaluated over repeated stratified cross-validation against eight baselines under one protocol, AdaStack attains the best mean rank (3.20 against 4.00) and the best mean macro F1 (0.787 against 0.784), and is the top model on three of five tasks; it significantly outperforms dynamic time warping on two tasks after Holm correction, while differences against the stronger baselines remain within the resolution of a 47-to-164-trial benchmark. We then close the loop in simulation, on a physics-based MuJoCo peg-in-hole cell. There the limiting factor turns out not to be perception: with a recovery policy written by hand, a spiral search that ignores the estimated state seats 78.5% of insertions while the learned monitor reaches 33.0% and even an oracle given the true fault state reaches 52.0%. Optimizing the twelve free policy parameters against a cost over seating, contact force, cycle time and damage, with disjoint optimization and evaluation seeds, yields in this simulated cell a controller that matches the search’s seating rate (76.5% against 78.5%) while cutting mean peak contact force by 46% (17.2 N against 31.6 N) using the learned monitor rather than an oracle. The same gain, however, is available without the monitor: a spiral search whose own three parameters are optimized with the same budget seats 79.0% at 18.3 N, and it matches or beats the monitor-driven policy under four cost weightings and five of six shifted operating conditions. What paid was optimizing the reaction, not conditioning it on the estimated state. Effort spent on the recovery policy repaid far more than effort spent on monitoring architecture; whether this transfers to a physical cell remains to be tested.

10 October 2026 in Preprints.org
Verified
Preprint
Open access
PDF file
Xu RuiXu Rui
Xu Rui
YZYu Zheng
Yu Zheng
YKYanxi Kuang
Yanxi Kuang
ZRZixuan Ren
Zixuan Ren
XWXianni Wen
Xianni Wen
JHJingqi Hong
Jingqi Hong
County-level hurricane power outage risk prediction faces a mismatch in spatial support among hazard rasters, infrastructure exposure, and outage labels. Conventional county-mean aggregation can weaken the spatial association between local hazard peaks and transmission corridors. To address this issue, we propose the Raster-to-Grid Spatiotemporal GeoAI Network (R2G-STGeoNet). It first combines Sentinel-1 SAR, precipitation, wind fields, and geographic environmental information to encode multi-source hazard states. Raster-to-Grid Exposure Projection (R2GEP) then transforms continuous hazard fields into features including county exposure, along-line exposure, local peaks, and lengths of above-threshold exposure. A dual-relation spatiotemporal graph combining geographic adjacency and transmission-corridor relationships jointly predicts outage occurrence probability and outage rate. Using hurricanes as case studies, we evaluate matched-input baselines, structural ablations, cross-event transfer, and spatial holdout, accounting for missing labels and data latency. The full model achieves an AUROC of 0.913; replacing R2GEP with county means lowers AUROC to 0.867. With identical R2G inputs, the proposed model outperforms R2G-ST-GNN (AUROC = 0.895). The results indicate that spatial support transformation and infrastructure relation modeling improve county-level outage risk prediction under extreme weather.

Verified
Journal Article
Open access
PDF file
Xu RuiXu Rui
Xu Rui
YZYu Zheng
Yu Zheng
YKYanxi Kuang
Yanxi Kuang
ZRZixuan Ren
Zixuan Ren
MQMeiyi Qi
Meiyi Qi
JHJingqi Hong
Jingqi Hong
County-level hurricane power outage risk prediction faces a mismatch in spatial support among hazard rasters, infrastructure exposure, and outage labels. Conventional county-mean aggregation can weaken the spatial association between localized high hazard values and transmission corridors. To address this problem, this study proposes the Raster-to-Grid Spatiotemporal GeoAI Network (R2G-STGeoNet). The method first integrates Sentinel-1 SAR, precipitation, wind fields, and geographical information to encode multisource hazard states. Raster-to-Grid Exposure Projection (R2GEP) then transforms continuous hazard fields into features describing county exposure, along-line exposure, local high values, and above-threshold exposure length. A dual-relation spatiotemporal graph combining geographic adjacency and transmission corridor relationships jointly predicts outage occurrence probability and outage rate. Using hurricanes as case studies, we conduct common-input baseline comparisons, structural ablations, cross-event transfer, and spatial holdout validation while considering missing labels and data latency. The full model achieves an AUROC of 0.913 in retrospective reconstruction, while the availability-constrained one-hour-ahead evaluation yields an AUROC of 0.884, decreasing to 0.862 when dynamic inputs unavailable at issuance are masked. Replacing R2GEP with county means reduces AUROC to 0.867, and the proposed model outperforms R2G-ST-GNN (AUROC = 0.895) under identical R2G inputs. The results indicate that spatial support transformation and infrastructure-relation modeling can improve county-level outage risk prediction under extreme weather.