主讲人

（单位）

潘凯

(香港理工大学)

主持人

（单位）

李四杰、金子亮

（东南大学）

讲座时间

2026年9月18日10点00分

讲座地点

综合楼314

主讲人简介

Kai Pan is currently an Associate Professor in Operations Management at the Faculty of Business of The Hong Kong Polytechnic University (PolyU), the Director of the MSc Program in Operations Management (MScOM), and the Deputy Director of the Doctor of Business Management (DBM) Program. He received his Ph.D. degree from the University of Florida, USA, in 2016 and his Bachelor's degree from Zhejiang University, China, in 2010. Before joining PolyU in 2016, right after his Ph.D., he worked as a Research Scientist at Amazon (Seattle, Washington) on Supply Chain Optimization and as a Power System Engineer at GE Grid Solutions (Redmond, Washington) on Electricity Market Operations. His research interests include Stochastic and Discrete Optimization, Robust and Data-Driven Optimization, Dynamic Programming, and their applications in Energy Markets, Smart Cities, Supply Chain, Shared Mobility, Telecommunications, and Marketing. His research on these topics has been published in Operations Research, Manufacturing and Service Operations Management, INFORMS Journal on Computing, Production and Operations Management, IISE Transactions, European Journal of Operational Research, IEEE Transactions on Power Systems, Transportation Research Part B, etc. He was the first-place winner of the IISE Pritsker Doctoral Dissertation Award in 2017 and the awardee of the PolyU Young Innovative Researcher Award (YIRA) 2025. He serves as an Associate Editor for IISE Transactions, Decision Sciences, and Omega, and has served as a Secretary/Treasurer for the INFORMS Computing Society (ICS).

讲座内容摘要

We study approximation and reformulation techniques for problems with distributionally robust chance-constrained linear matrix inequalities (DRCCLMI) to overcome the computational challenges posed by multidimensional integration and nonconvexity of feasible sets. We consider the case where uncertain parameters are only partially described by support, mean, and covariance information. For a block-structured linear matrix inequality (LMI), a generalized Schur complement reduces the matrix chance constraint to a quadratic chance constraint. When no support restriction is imposed, this reduction and Conditional Value-at-Risk (CVaR) reformulation yield an exact finite-dimensional semidefinite programming (SDP). When explicit support information is present, we derive a semi-infinite inner approximation. For ellipsoidal supports, the resulting semi-infinite quadratic constraints are converted into SDP constraints via the S-lemma. For polyhedral supports, we construct a reformulation--linearization technique (RLT)-based SDP relaxation that is tighter than the corresponding CVaR-based approximation. For general DRCCLMI, we use a minimum-eigenvalue loss to derive a CVaR-based approximation and model the resulting recourse function using piecewise linear decision rules (PLDR). We further address the remaining robust LMI by deriving reformulations for different support geometries and developing a delayed constraint generation algorithm with separation routines for ellipsoidal and polyhedral supports. The effectiveness of these techniques is demonstrated through numerical studies on two applications: truss topology design and calibration problems.