基于数据驱动的发展式头脑风暴优化算法研究

In general, traditional swarm intelligence methods have used the same operators on different problems. The abandonment of problem characteristics has led the algorithm's poor generalization ability. It is a dynamical and developmental process for swarm intelligence algorithms solving problems. With the enhancement of learning ability and developmental ability, more problems could be solved more efficiently..In this project, the dynamic search process of the BSO optimization algorithms is analyzed based on the data analytics technology. Data analytics methods are proposed based on the BSO optimization algorithms. The study primarily analyses the relations among search status and population diversities of BSO algorithms and the different types of optimized problems. BSO algorithms could dynamically change its attributes based on the relations among algorithms and various problems. This is followed by interpreting the solution information on solving various problems as a collection of data. Such a search process could be analyzed and the performance could be enhanced via the data analytics on the data collection. Finally, data analytics problems would be represented as dynamic, multiobjective and large-scale optimization problems. New data analytics method will be further applied to the dynamic protein network data analysis and forecast problems. This study will benefit for identifying the relations among BSO algorithm and various problems, enhancing the efficiency of BSO algorithm on solving various problems, and solving the search based data analytics problems via BSO based data analysis methods.

传统的群体智能优化算法使用了相同的优化算子求解不同类型优化问题，缺乏问题特征的学习过程，存在泛化能力不足的问题。群体智能优化算法的优化过程是动态的、发展式的，改进算法的学习能力和发展能力将提高算法的性能和适用范围。本项目将研究基于数据驱动的搜索策略和具有通用性的数据分析问题建模过程，设计基于数据驱动的发展式头脑风暴优化算法；将数据分析问题建模为优化问题，利用发展式优化算法进行求解，包括：（1）分析求解问题时算法搜索状态和群体多样性变化，建立优化算子和求解状态的关联，使算法能针对问题进行自适应调整；（2）将算法求解过程中的解集合信息视为数据，用数据分析方法对其进行分析，揭示算法运行机理；（3）将数据分析问题建模为动态多目标大规模优化问题，利用优化算法进行求解，并以实际数据分析问题验证。通过本项目，将建立头脑风暴优化算法求解机制，设计基于数据驱动的发展式优化算法，改进算法求解性能和适用范围。