可融合偏好的大规模进化优化算法研究

Large-scale optimization problems are very common in various real-world applications. Although the class of evolutionary optimization algorithms is one of the most effective approaches for solving complex optimization problems, most existing algorithms are only applicable to small-scale problems (e.g. with smaller than 100 decision variables), and thus cannot fulfill the requirements of many important applications. In addition, since the solution space of a large-scale problem is often very large, it is not realistic to always pursue its global optimum. So providing a satisfying solution based on user preferences is often a better choice. The preferences of different users, or even different periods of the same user, may be quite different, so introducing a mechanism, which can incorporate different user preferences with the optimization algorithms, becomes very important. To address the above issues, this project will mainly focus on new ideas for solving large-scale optimization problems, methods to extract and incorporate user preferences, and finally delivering preference incorporated large-scale evolutionary optimization algorithms. The performance of the design algorithm will be evaluated and analyzed on both standard benchmarks and a real-world application. The key acquisitions of this project are potential to break the bottleneck of evolutionary optimization algorithms on large-scale problems, and also provide a way to incorporate user preference to deliver satisfying solutions.

实际应用中存在着大量的大规模优化问题，进化优化算法虽然是目前求解该类复杂优化问题的有效方法之一，但现存算法一般仍只适用于决策变量比较少（如少于100个）的小规模问题，无法满足应用需求。此外，由于大规模优化问题解空间非常庞大，仍苛求其全局最优解显得很不现实，根据用户偏好提供满意解往往是更有效的解决方案，而不同用户甚至同一用户在不同时期对问题解的偏好很可能存在较大差别，因此在优化算法中加入可融合偏好的机制显得非常重要。针对上述问题，本项目主要研究面向大规模优化问题的新求解思路，以及优化过程中提取和融合偏好的有效方法，进而设计一种可融合偏好的大规模进化优化算法，并采用标准测试问题集和实际应用问题对其性能进行测试与分析。本项目关键技术的攻克将有望使进化优化算法在一定程度上突破求解大规模优化问题的瓶颈，并可方便地融合用户偏好提供满意解。

实际应用中存在着大量的大规模优化问题，进化优化算法虽然是目前求解复杂优化问题的有效方法之一，但现存算法一般仍只适用于决策变量比较少（如少于100个）的小规模问题，无法满足应用需求。此外，由于大规模优化问题解空间非常庞大，仍苛求其全局最优解显得很不现实，根据用户偏好提供满意解往往是更有效的解决方案，而不同用户甚至同一用户在不同时期对问题解的偏好很可能存在较大差别，因此在优化算法中加入可融合用户偏好的机制显得非常重要。针对上述问题，本项目主要研究面向大规模优化问题的新求解思路，以及优化过程中提取和融合用户偏好的有效方法，进而设计一种用户偏好可配置型大规模进化优化算法，并采用标准测试问题集和实际应用问题对其性能进行评测与分析。本项目研究的突破将有望使进化优化算法在一定程度上突破求解大规模优化问题的瓶颈，并使其针对用户偏好具有良好的可配置性。