具有稳定性保证的在线学习:一种基于记忆的实时模型预测控制器
最优化与控制
2020-09-23 v5
摘要
我们提出并分析了一种实时模型预测控制(MPC)方案,该方案利用存储数据通过在线学习值函数并以稳定性保证提升其性能。针对线性和非线性系统,给出了一种学习方法,其利用代价函数的基本解析性质,并被证明可在闭环状态轨迹的极限集上学习到MPC控制律与值函数。核心思想是基于历史数据生成智能热启动,以改善未来数据点从而改善未来的热启动。我们表明这些热启动渐近精确并收敛于MPC优化问题的解。由此,由实时要求引起的所施加控制输入的次优性随时间消失。仿真示例表明,现有实时MPC方案可通过存储数据与所提学习方案得以改进。
引用
@article{arxiv.1812.09582,
title = {Online learning with stability guarantees: A memory-based real-time model predictive controller},
author = {Lukas Schwenkel and Meriem Gharbi and Sebastian Trimpe and Christian Ebenbauer},
journal= {arXiv preprint arXiv:1812.09582},
year = {2020}
}
备注
This article is an extended version of the paper "Online learning with stability guarantees: A memory-based warm starting for real-time MPC" published in Automatica, Volume 122, 109247, 2020, including all proofs, an application example, and a detailed description of the used algorithm