求解大规模数据分析中复杂优化问题的演化算法研究

Aiming at the complex optimization problems in large-scale data analytics, in this project, we investigate the resolution strategies based on evolutionary algorithms, the theoretical fundation of the algorithms, as well as the applications in large-scale data analytics. Firstly, based on our previously proposed social algorithm (SA) and random drift particle swarm optimization (RDPSO), a graph partition method for objective function decomposition and a multi-tier cooprative coevolution mechanism are proposed to tackle the high-dimensional optimization problems, and the controlling strategies for diversity decline speed and diversity maintenance are invented for dynamic and multi-objective optimization problems. Next, the convergence, convergence rates, time complexities and population diversity of the algorithms are analyzed in order to provided the theortical basis for solving high-dimensional dynamical multi-objective optimization problems by the algorithms. Finally, the above mentioned algorithms and resolution strategies are applied to the paramter optimization for large-scale models, feature selection for high-dimensional data, time series analysis for stream data. The achivement of this project will provide effective approaches and theoretical guidelines for solving complex optimization problems in large-scale data analytics, so that it will be of great theoretical and practical significance.

本项目针对大规模数据分析中的复杂优化问题，研究基于演化算法的求解策略、算法理论基础、算法策略在大规模数据分析中的应用。首先，基于课题组前期提出的社会算法(Social Algorithm, SA)和随机漂移粒子群优化(Random Drift Particle Swarm, RDPSO)算法，设计解决高维优化问题的基于图划分的目标函数分解策略和多层次多群体的合作协同演化机制，研究解决动态多目标优化问题的多样性减小速度控制和多样性保持策略。其次，分析SA和RDPSO算法的收敛性、收敛率、复杂性和多样性，为算法解决高维、动态和多目标优化问题提供理论依据。最后，针对大规模模型参数优化、高维数据特征选择和流数据的时间序列分析等问题，研究算法及相关策略的在大规模数据分析中实际应用。本项目的研究成果为解决大规模数据分析中的复杂优化问题提供有效方法和理论指导，具有重要的学术意义和应用价值。

本项目针对大规模数据分析中的复杂优化问题，研究基于演化算法的求解策略、算法理论基础、算法策略在大规模数据分析中的应用。首先，基于课题组前期提出的l量子行为粒子群优化（Quantum-behaved Particle Swarm Optimization, QPSO)算法和随机漂移粒子群优化(Random Drift Particle Swarm, RDPSO)算法，设计了解决高维优化问题的基于差分分解的目标函数分解策略和多层次多群体的合作协同演化机制，研究 解决动态多目标优化问题的多样性减小速度控制和多样性保持策略。其次，分析QPSO和RDPSO算法的收敛性、收敛率、复杂性和多样性，为算法解决高维、动态和多目标优化问题提供理论依据。最后，针对大规模模型参数优化及其在计算机视觉中的应用、基于多目标优化的数据分析、生物分子对接等复杂优化问题，研 究算法及相关策略的在实际复杂优化问题应用。本项目的研究成果为解决大规模数据分析、实际工程中的复杂优化问题提供有效方法和理论指导，具有重要的学术意义和应用价值。