Wang Yi

Mr. Wang Yi

orcid

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
24 June 2024 in Transportation Research Record: Journal of the Transportation Research Board
Verified
Journal Article
Closed access
QLQin Luo
Qin Luo
SHShan Huang
Shan Huang
WLWei Li
Wei Li
WYWang Yi
Wang Yi
CZCuifeng Zeng
Cuifeng Zeng
JCJingjing Chen
Jingjing Chen
Optimizing the maintenance scheduling of metro systems is a crucial task that necessitates meticulous coordination of labor, equipment, and workspaces to ensure optimal system performance and safety. A mathematical model and a two-stage teaching-learning-based optimization (TLBO)-resource operators crossover (ROC) algorithm are proposed aiming at optimizing the scheduling of maintenance tasks for metro systems. The mathematical model focuses on minimizing the makespan, which represents the total duration or time required to complete a set of tasks or activities within a project. In addition, it takes into account the need to balance the load on labor and workspaces, considering environmental constraints, limited resources, and strict scheduling requirements. A two-stage TLBO-ROC algorithm is specifically designed to enhance the scheduling process. It achieves this by iteratively updating the local best individual matrix, dividing it into groups, and adjusting the resource allocation. This algorithm effectively reduces the makespan while also achieving improved balance in the workspace load. The model and algorithm are tested on the Shenzhen metro system. Experimental results demonstrate that our proposed approach significantly reduces the makespan. In comparison to manual scheduling plans, the algorithm achieved a remarkable 28.06% reduction in the makespan. Moreover, when compared to benchmark algorithms, our proposed algorithm not only improves the makespan but also ensures more equitable occupation of workspaces by maintaining a similar balance in labor load.

20 May 2026 in Reliability Engineering & System Safety
Verified
Journal Article
Closed access
SHShan Huang
Shan Huang
WYWang Yi
Wang Yi
LLLishuai Li
Lishuai Li
BLBo Liang
Bo Liang
LQLuo Qin
Luo Qin


29 September 2026 in Journal of Cleaner Production
Verified
Journal Article
Closed access
JHJunsheng Huang
Junsheng Huang
KHKang Huang
Kang Huang
BMBaohua Mao
Baohua Mao
WYWang Yi
Wang Yi
YLYinzheng Lin
Yinzheng Lin

2 April 2026 in SSRN Electronic Journal
Verified
Preprint
Open access
SHShan Huang
Shan Huang
YYYongchao Ye
Yongchao Ye
ANAvraham Natan
Avraham Natan
WYWang Yi
Wang Yi
WPWeijian Pan
Weijian Pan
Lishuai LiLishuai Li
Lishuai Li
Real-time train timetable rescheduling is constrained by a fundamental trade-off between solution optimality and computational speed. This work introduces an Large Language Model (LLM) Agent-Guided Optimization Framework that produces near-optimal solutions with sub-second latency. The framework integrates a LLM agent with a mathematical solver to overcome the documented limitations of existing methods. Conventional optimization methods, while guaranteeing global optimality, are too computationally intensive for real-time applications. Conversely, heuristics often fail to generalize to new network topologies, reinforcement learning can produce sub-optimal results, and direct LLM applications are limited by physical constraint violations and high inference latency. Our methodology involves fine-tuning an LLM agent on a dataset of high-quality solving traces from an exact solver. This process enables the agent to learn effective solution strategies. During inference, the agent generates strategic configurations—such as warm starts, variable fixings, and branching priorities—which are fed to a Mixed-Integer Linear Programming (MILP) solver. These configurations guide the solver, significantly pruning its search space and accelerating convergence. Experimental results on complex network scenarios show our framework achieves a computational speedup of up to 110-fold compared to a standard MILP baseline, reducing decision latency to under one second. A key finding is that the framework's inference time is largely decoupled from problem size and complexity. This allows it to maintain near-optimal solution quality in large-scale scenarios where the performance of the baseline solver degrades significantly.

19 June 2026 in Applied Sciences
Verified
Journal Article
Open access
PDF file
CYChunlong Yu
Chunlong Yu
YPYi Peng
Yi Peng
KLKunyan Li
Kunyan Li
JGJianyu Guo
Jianyu Guo
WYWang Yi
Wang Yi
JCJingjing Chen
Jingjing Chen
Equipment fault prediction in maglev train systems poses substantial challenges: fault events are inherently rare, class distributions are severely imbalanced, and individual equipment units are subject to complex spatial and functional couplings that single-device statistical approaches fundamentally cannot capture. To address these challenges, this study proposes a Knowledge Graph-driven Graph Neural Network (KG-GNN) framework. A fault knowledge graph encompassing equipment, fault, temporal, and environmental entities is constructed to unify multi-source maintenance data. Graph connectivity is established via three spatial relation types (co-location, co-zone, and co-level), with edge weights derived from Laplacian-smoothed Lift scores quantifying fault co-occurrence strength. A two-layer GATv2Conv-based graph attention network is designed: the first layer employs four-head attention with explicit edge-weight integration to capture heterogeneous neighborhood influences, while the second layer produces compact node embeddings via single-head attention. A Top-20 sparsification strategy suppresses weak-association noise, and training under severe class imbalance is stabilized through Focal Loss and F2-Score-guided early stopping. On the test set, the proposed method achieves an F2-Score of 0.5703, Recall of 0.6825, and AUC-ROC of 0.9329 (single-run evaluation); multi-seed evaluation (5 seeds) yields F2 = 0.5645 ± 0.0035, Recall = 0.6789 ± 0.0095, and AUC-ROC = 0.9298 ± 0.0026, outperforming the MLP baseline by 18.3% in F2-Score and substantially exceeding GCN (F2 = 0.1476 ± 0.0176) and GATConv (F2 = 0.4284 ± 0.0097). Ablation studies confirm the individual contributions of authentic graph topology, precise edge weighting, and graph sparsification to overall performance.