BMI Global Optimization with Beowulf Cluster and Simultaneous Optimization of Structure-Control

1.基于Beowulf集群的BMI优化研究(1)构建了一台Beowulf集群计算机，使我们在104个CPU的情况下可以达到2[十]的精度。(2)在BMI全局优化的分支定界方法上，我们提出了一种新的分支操作，考虑了复杂的变维问题。(3)针对BMI分支定界方法的并行化，我们开发了实现变计算粒度的并行化算法。在BMI优化的实数编码遗传算法方面，引入了一种降阶个体表达式和一种基于LMI优化的个体评价方法。与传统的基于特征值的方法相比，该方法能够更准确地评估个体。(5)为了减少改进方法(4)的计算负担，我们还开发了一种充分利用Beowulf集群计算能力的并行化算法。基于BMI优化方法的并行优化研究(1)由于链路结构的复杂性，并行链路系统是并行优化的最佳应用之一。并联系统的性能既取决于控制器，也取决于并联结构。针对冗余并联系统，提出了一种基于LMI优化的最优功率分配控制方法。从实用的角度来看，建模和参数整定对于同时优化设计都是非常重要的。在此背景下，我们提出了一种基于数据的控制器设计方法，该方法基于无误控制技术和支持向量机。该方法阐明了基于数据的同时优化方法的可能性。

1. Research for BMI optimization with Beowulf cluster(1) A Beowulf cluster computer that enables us to achieve 2[decade] accuracy with 104 CPUs is constructed.(2) On branch and bound method for BMI global optimization, we developed a novel branching operation considering complicating variable dimension which is important on practical problems.(3) For the parallelization of the BMI branch and bound method, we developed parallelized algorithms that realize variable computational granularity. The developed algorithms of the BMI branch and bound are implemented in Beowulf cluster.(4) On real-coded genetic algorithm for BMI optimization, we introduced a reduced-order individual expression and an individual evaluation method using LMI optimization. The developed method enables us to more accurately evaluate individuals compared to conventional eigenvalue-based methods.(5) In order to reduce the computation burden of the developed method (4), we also developed a parallelization algorithm that fully utilizes the computational power of Beowulf cluster.2. Research for simultaneous optimization based on BMI optimization method(1) Parallel-link system is one of the best application of simultaneous optimization due to the complexity of the link structure. The performance of the parallel-link system depends on both controller and link-structure. For redundant parallel-link system, we developed a optimal power distribution control with LMI optimization. However, the globally optimal link-structure and globally optimal control row are not yet obtained.(2) In the practical view point, modeling and parameter tuning are both important for simultaneous optimization design. In the context, we developed a data-based controller design method based on unfalsified control technique and support vector machine. The method clarified the possibility of the data-based simultaneous optimization method.