基于受限贝叶斯优化的控制器参数整定

Controller parameters tuning with constrained Bayesian optimization

中文关键词  参数自整定  安全优化  高斯过程  轨迹跟踪  粒子群优化

英文关键词  automatic parameter tuning  safe optimization  Gaussian process  trajectory tracking  particle swarms optimization

中文摘要

      参数化固定结构前馈控制方法可以有效地将前馈力控制问题转化为控制器参数自整定问题. 但控制器参数整定问题往往非常耗时, 且传统的贝叶斯优化容易在参数整定过程中出现系统失稳等不安全的情况, 针对此问题, 本文提出了一种基于受限目标函数的贝叶斯安全优化方法. 首先通过高斯过程获得调参对象的代理模型, 然后通过在代理模型上添加预先设置的安全阈值, 借助粒子群算法来完成贝叶斯优化的评估点的求解, 从而实现控制力的更新, 最终通过迭代的方法得到最优参数, 并且在对应参数的前馈控制下可以实现满足安全约束条件下的运动控制系统轨迹最优跟踪性能. 实验结果验证了所提出的算法能够实现安全约束条件下最优点到点轨迹跟踪性能.

英文摘要

      The feedforward force control problem can be transformed into a controller parameter self-tuning problem by using parameterized fixed-structure feedforward control methods. But the controller parameter tuning is often timeconsuming, and the traditional Bayesian optimization is prone to unsafe situations such as system instability during the parameter tuning processes. To solve this problem, a Bayesian safe optimization incorporating objective function constraints is proposed. Firstly, the agent model of the tuning parameters is obtained by the Gaussian process. Then the safe evaluation point is determined in Bayesian optimization by incorporating a pre-set safety threshold and leveraging the particle swarm algorithm, which enabling control force updating. Eventually, the optimal parameters are obtained by an iterative process, and the optimal trajectory tracking performance of the motion control system can be achieved by the feedforward controller with the corresponding optimal parameters while adhering to safety constraints. The experimental results verify that the proposed algorithm can achieve the optimal point-to-point trajectory tracking performance under the safety constraints.