基于变复杂度和GPU并行的高效多响应稳健优化算法研究

Robust optimization method can effectively improve the quality of products.However,traditional robust optimization methods are mainly concentrated in single response robust optimization,and their efficiency is very low,which restricts their application in the design of complex products.There are some common basis problems to carry out the research on multi-response robust optimization method including multiple responses conflicting with each other and to improve the efficiency and accurancy of robust optimization method needed to be addressed urgent in engineering optimization design field.The project utilizes the advantage of the hybrid surrogate technique,variable fidelity technique and sequential approximation technique to present an adaptive variable fidelity hybrid surrogate (AVFHS)modeling method. In addition,the project investigaes the parallel solving mechanism of the artificial bee colony algorithm (ABC) and develops a parallel accelerating multiobjective artificial bee colony algorithm based GPU (GPU-MOABC).Finally,the project establishes a multi-response robust optimization mathematical model considering parameter fluctuations and multiple performance characteristics conflicting with each other, explores the decoupling technique and solving method of multi-response robust optimization mathematical model, and carry out the parallelization on the GPU platform. The crashworthiness optimization design of aluminum foam filled thin-walled structure is demonstrated to verify the effectiveness of the method presented. The research resutls provide a high efficient scheme to substantial improve the design level for complex products.

稳健优化方法能有效提高产品的质量。然而，传统的稳健优化方法主要集中在单响应稳健优化研究方面，并且效率低下，限制了其在复杂产品设计中的应用。开展多个响应互相冲突的多响应稳健优化方法研究并且提高稳健优化方法的求解效率和精度是工程优化设计领域迫切需要解决的共性基础问题。本项目充分利用混合代理模型技术、变复杂度技术和序列近似技术的优点，提出一种自适应变复杂度混合代理模型(AVFHS)的建模方法；研究人工蜜蜂算法的并行求解机理，开发基于GPU并行加速的多目标人工蜜蜂算法(GPU-MOABC)；建立考虑参数波动和多个质量特性互相冲突的多响应稳健优化数学模型，探讨多响应稳健优化模型的解耦技术和求解方法，并在GPU平台上实现并行化。以泡沫铝填充薄壁结构的耐撞性优化设计为例，验证提出的方法的有效性。该项目的研究成果为大幅提高复杂产品的设计水平提供一个高效的解决方案。

稳健优化方法能有效提高产品的质量。然而，传统的稳健优化方法主要集中在单响应稳健优化研究方面，并且效率低下，限制了其在复杂产品设计中的应用。开展多个响应互相冲突的多响应稳健优化方法研究并且提高稳健优化方法的求解效率和精度是工程优化设计领域迫切需要解决的共性基础问题。本项目充分利用混合代理模型技术、变复杂度技术和序列近似技术的优点，提出一种自适应变复杂度混合代理模型(AVFHS)的建模方法；研究人工蜜蜂算法的并行求解机理，开发基于GPU 并行加速的多目标人工蜜蜂算法(GPU-MOABC)；建立考虑参数波动和多个质量特性互相冲突的多响应稳健优化数学模型，探讨多响应稳健优化模型的解耦技术和求解方法，并在GPU 平台上实现并行化。以泡沫铝填充薄壁结构的耐撞性优化设计为例，验证提出的方法的有效性。该项目的研究成果为大幅提高复杂产品的设计水平提供一个高效的解决方案。