时间：2026-01-06 10:00
举办地点：产教融合大楼209
内容：
In real-time applications such as embedded systems and robotics, the online computation of Model Predictive Control (MPC) is often limited by solving multiparametric quadratic programming (mpQP) instances. Existing methods handle their constraints, incurring high computational cost, while many constraints are actually redundant and their removal does not change the optimal solution. Using solved mpQP instances, we propose a safe and optimality certifiable strategy for removing redundant constraints, thereby substantially reducing the scale of the mpQP instances to be solved. In particular, we prove that the number of constraints of mpQP of MPC reduces to zero for the constrained linear stabilization problem. Furthermore, we explicitly quantify the complexity of a ReLU network-based MPC policy to ensure the closed-loop performance in terms of network depth and width parameters. Both theoretical analysis and experiments confirm that the method maintains control performance while achieving computational speedup, offering an advanced solution for resource-constrained platforms.
主办单位：
电气学院
主讲人：
游科友，清华大学自动化系长聘教授、博士生导师，国家杰出青年科学基金获得者。2007年获中山大学统计科学学士学位，同年8月至2012年6月于新加坡南洋理工大学电气与电子工程学院攻读博士学位并从事博士后研究。2012年7月起任教于清华大学自动化系，曾受邀访问意大利都灵理工大学、澳大利亚墨尔本大学、香港科技大学等高校。长期致力于复杂网络化系统的学习、优化与控制研究。现任Automatica、IEEE Transactions on Control of Network Systems等国际期刊副编委，主持国家自然科学基金杰青与重点项目、联合基金项目、科技创新2030—“新一代人工智能”重大项目（青年）等。获中国自动化学会自然科学一等奖（排第一）、亚洲控制学会Temasek青年教育者奖、关肇直最佳论文奖，参与获北京市、教育部、自然资源部等省部级一等奖3项。指导博士生获 IFAC Young Author Award、中国自动化学会优秀博士学位论文奖、国家博士后创新人才支持计划等荣誉。
