非线性模型预测控制的快速计算方法研究及应用

The online optimization calculation of nonlinear model predictive control (NMPC) is the key bottleneck restricting its application in fast systems. This project focuses on the efficient model predictive control algorithms for applications with nonlinear systems, to deal with the limitations of current MPC methods including weak real-time performance and high requirement on hardware platform. The research mainly contains: (1) design the parallel computing structure of NMPC based on the multiresolution theory, which can enhance the computing performance through parallel computing of the predictive model; (2) design the simplified solution methods of the nonlinear optimization problem. In order to reduce the computational complexity, the nonlinear optimization problem is transformed to the algebraic equations problem to be solved; (3) explore the performance analysis of the computing algorithms of NMPC to guarantee the system performance through error analysis; (4) study the hardware implementation scheme of the NMPC controller based on the embedded hardware platform. Finally, evaluate the efficient nonlinear model predictive control algorithms with the application to vehicle control system. From this study, we can enhance the real-time computational performance of MPC and meet the challenges in the practical application of MPC. Moreover, this research project will provide not only new method for MPC research, but also technical basis for promoting the popularization and application of MPC.

非线性模型预测控制的在线优化计算量大是制约其在快速系统中应用的关键瓶颈。本项目围绕非线性预测控制的快速计算方法展开研究，解决现有预测控制算法实时性差、对硬件实施平台要求高的不足。具体内容包括：(1)研究非线性预测控制的多分辨率设计方法，通过预测模型的并行计算，提高算法计算性能；(2)研究非线性优化问题的简化求解算法，将非线性优化问题转换成代数方程组问题求解，简化求解过程，降低计算复杂度；(3)探索非线性预测控制计算算法的性能分析，通过算法误差分析保证系统性能；(4)研究非线性预测控制器的嵌入式硬件实现方案。最后，在汽车控制系统中进行非线性预测控制快速计算方法的应用验证。本项目的研究不仅可有效提高非线性预测控制的在线计算性能，解决实际应用研究的迫切需求，同时为预测控制算法的研究提供新方法，为其应用推广提供技术基础。

越来越多的快速动态系统迫切需要应用模型预测控制方法来处理其复杂约束优化问题和提高控制性能，然而非线性模型预测控制需要在线求解约束优化问题，在线计算负担大、实施平台要求高，制约了其在快速动态系统中的应用。本项目面向汽车控制等快速系统应用，旨在研究非线性预测控制快速计算方法和嵌入式硬件实现方案。主要工作包括：建立了基于时间变量和尺度变量描述的系统多分辨率模型，设计了多分辨率模型预测控制器，提出了基于干扰观测器的模型预测控制方法，给出了非线性预测控制并行牛顿求解方法、融合车辆机理知识的极值原理快速求解方法，提出了预测控制器的FPGA异构计算硬件并行实现方案。在车辆横摆稳定控制、车辆路径跟踪控制、柴油机发动机控制等应用方面取得了一些成果，克服了在汽车控制等快速动态系统中嵌入式实现的本质困难。