基于演化胞映射的复杂控制系统的高维多目标优化设计

Optimal design of complex control system usually involves multiple objectives. These objectives are often mutually conflicting. Therefore, it is quite necessary to perform multi-objective optimization for complex control system design. This project will combine the evolutionary algorithm and cell mapping method for the development of a multi-objective searching algorithm suitable for high dimensional problems. Theoretical aspects of the proposed algorithm such as convergence and global coverage will be studied. The algorithm will be implemented via graphic processing units (GPU) in parallel manner for speed up purposes. Such attempt will provide an economic and reliable technical support and computational platform for solving large scale engineering optimization problems. The algorithm will be applied to design complex feedback control system with nonlinearity, time delay, uncertainty and randomness. Control strategy involved in multi-objective optimal design include linear/nonlinear PID, sliding mode control, backstepping and adaptive control. Time domain indices are considered to evaluate the stability and robustness of the controllers. The optimal controllers will be further validated through the trajectory tracking of autonomous quadrotor drones under complicated terrain.

复杂控制系统的优化设计往往涉及多个指标。这些指标通常相互制约、相互矛盾。因此针对复杂控制系统的多目标优化设计十分必要。本项目将通过结合演化算法和胞映射算法，设计开发出一套适合高维多目标优化的搜索算法，并对算法的收敛性和全局性进行理论论证。通过显卡计算单元的并行实现，对算法进行提速，从而为解决大型工程优化问题提供可靠廉价的技术支持和计算平台。应用所提算法，我们将对含非线性、时滞、不确定性、随机性等因素的复杂系统进行闭环控制的多目标优化设计。控制策略包括线性/非线性PID、滑模控制、反演控制和自适应控制。通过考虑时域指标来衡量控制系统的稳定性和鲁棒性。所设计的多目标优化控制器将通过四旋翼无人机在实地复杂地形环境下的路径追踪试飞来进行综合考察和评估。

复杂控制系统的优化设计往往涉及多个时域（频域）指标，而这些指标通常都是相互矛盾和相互制约的。本项目针对复杂非线性控制系统，提出一种适合并行计算的多目标优化算法，并将其应用到解决非线性控制系统的多目标优化问题中。项目在执行期内主要从算法开发、理论分析、控制器设计和实验验证四部分展开研究。整个课题在方法论和应用上对多目标优化问题进行了深入的研究，并将该方法应用在了以控制器多目标优化设计为代表的的诸多领域。主要研究内容及成果概括如下：（1）基于演化胞映射算法的多目标优化控制问题研究；（2）基于演化胞映射算法智能交通动力学问题的研究；（3）基于深度学习算法的智能预测研究；（4）基于自适应动态规划的主动质量阻尼系统振动控制；（5）高维多目标结构优化设计；（6）基于演化胞映射算法的全局路径规划问题研究；（7）随机系统动力学分析相关方面研究。