Wei Tai

Dr. Wei Tai


Research Focus & Metrics

Representative Publications

View All Publications
26 October 2026 in Preprints.org
Verified
Preprint
Open access
PDF file
CCCen Chen
Cen Chen
JLJinghong Lan
Jinghong Lan
NLNuannuan Li
Nuannuan Li
JCJunfei Cai
Junfei Cai
WTWei Tai
Wei Tai
QWQi Wang
Qi Wang
A valid credential can allow a compromised application to sustain harmful control actions after zero-trust admission. We develop stateful capability leases that account for command amplitude and duration, preserve consumed authority across renewals, and transfer control to a protected fallback at expiration. The analysis bounds cumulative command exposure and stale authority, relates review capacity to authorization freshness, and provides a conditional physical-invariance result. Nonlinear four-region feeder experiments examine an in-memory ledger with an affine physical filter. Over twelve pairs of six-hour campaigns, the local and matched central ledgers contained all unauthorized-power goals given at a budget chosen; risk-adaptive and short-lease approaches contained only three. Median aligned direction attack minus clean energy expenditure dropped from 2.3965 to 0.1231 kWh as compared with role-based access. The ledger containment range is [0.608, 1] at 95% confidence level. Local placement modestly improved partition service, while regional withdrawal restricted benign service. Higher budgets made possible successful attacks despite proper accounting and nonlinear stresses demonstrated remaining shared-power contract violations. In a discrete-event authorization experiment, synchronization increased the p99 reauthentication delay from 0.249 to 3.645 s under equivalent light cost hypothesis.

4 November 2026 in Modelling
Verified
Journal Article
Open access
PDF file
WTWei Tai
Wei Tai
JCJinchang Chen
Jinchang Chen
JZJianxin Zhang
Jianxin Zhang
SLShiyang Li
Shiyang Li
QWQi Wang
Qi Wang
LTLian Tu
Lian Tu
Online power-system defense increasingly depends on automated assessment, strategy generation, and real-time implementation. However, the accuracy of a single model or controller does not establish the trustworthiness of the complete closed loop. This paper proposes a mission-level framework for closed-loop trustworthiness under physical, informational, and temporal uncertainty. Trustworthiness is defined as the lowest finite-horizon mission-success probability over all probability laws consistent with available evidence. The resulting trustworthiness viability domain contains the joint physical, evidence, dependence, and response-time states from which a required trustworthiness level remains achievable. A recursive characterization supports model-conditional certification and boundary warning. Additional evidence can reduce maintainable trustworthiness when its acquisition delay removes feasible recovery actions. Safe abnormality identifiability requires diagnostic actions to remain inside the viability domain, while a counterfactual deficit measure locates abnormalities with the largest mission-level effect. Conditions are also derived for heterogeneous redundancy and certified fallback. On a modified Reliability Test System–Grid Modernization Lab Consortium (RTS-GMLC) system, 14.3% of test states were physically secure but were not certified at the required trustworthiness level. Under the scenario-based model, 94.6% of abnormal cases met the 0.95 trustworthiness threshold with runtime selection. Texas2k tests quantified the accuracy–runtime tradeoff at a larger network scale.