协同生态粒子群计算模型及动态优化方法研究

Various dynamic optimization problems in real world should been solved validly, so it's necessary to develop swarm intelligence computation with good dynamic optimization performance. Since hybrid ecology systems in nature can respond to the change of the environment with more self-organization, self-learning and self-adaptation, its cooperative evolutionary mechanisms can provide us robust intelligent models. Following the intelligent behavior modes of the hybrid ecology systems, such as prey, competition, cooperation or intergrowth, cooperative ecology-particle-swarm-optimization models are developed in the project firstly. Secondly, valid feedback and "Baldwin-based" learning mechanism are introduced to improve the performance of the cooperative models. Moveover, many researches are done based on system theory, information theory and cybernatics, including the emergence mechanism of swarm intelligence and adaptive control methods; at the same time, the convergent analysis is made based on the support set, stochastics optimization theory and martingale theory. Finally, the cooperative models are applied to solve the complex optimizations in various dynamic environment, some valid monitoring and responding strategy are developed to improve the dynamic performances. In conclusion, the researches in the project are valuable and can greatly contribute to the design of swarm-intelligence-based model, adaptive control of the optimization process and the solutions of complex dynamic problems in real world.

现实世界中普遍存在的复杂动态优化问题，要求群智能计算具有高效的动态优化性能。而自然界中的混合生态系统，对生态环境的变化具有更强的自组织性、自学习性和自适应性，其丰富的协同进化机制能够为群智能计算提供鲁棒的智能模拟原型。本项目首先对混合生态系统的共生、捕食、竞争和协作等智能行为模式进行分析，在此基础上设计高效的协同生态粒子群计算模型；其次引入有效的反馈控制机制和Baldwin学习机制，提高协同粒子群模型的优化性能，同时基于系统论、信息论和控制论的思想，对协同粒子群模型的群智能涌现机理以及自适应控制方法进行研究，并采用支撑集、随机优化、鞅理论等数学工具对所建立的模型进行理论分析；最后基于动态环境优化问题对其动态性能加以分析和改进，设计高效的监测和响应策略。本项目的研究内容，能够为群智能计算方法的模型设计、优化过程的自适应控制以及实际工程中复杂动态优化问题的求解提供有益的思路和技术工具。

粒子群算法是一种典型的群智能计算方法，被证明是求解大规模分布式复杂问题的一种有效工具，已被广泛应用于各种实际工程领域。然而，随着求解问题复杂性的增加，现有粒子群算法模型面临着早熟、收敛效率低、全局收敛性能差等问题的挑战。针对粒子群优化算法所存在的缺陷和局限性，本项目从方法、模型以及工程应用等几个方面进行了系列研究。首先，设计了基于混合生态群体的协同粒子群优化模型，给出了其形式化描述，并对其优化性能进行了实验和理论分析；其次，基于混沌理论，构造了混沌粒子群优化模型，通过复杂多模态函数优化以及过渡金属配合物磁参数拟合优化问题，验证了该模型的有效性；在此基础上，将混沌粒子群优化模型拓展为多种群协同模型成功求解蛋白质结构预测问题；接着，针对多运动目标跟踪问题，设计互异多种群协同粒子群优化模型，开发互异多子群搜索策略，j基于共享信息对进化模型进行了改进；最后将粒子群算法推广至水质有机污染三维荧光分析中的组合优化建模、小水电站调度优化等实际工程问题中。系列研究结果表明：本项目内容有效丰富了粒子群算法的理论和应用研究，对算法全局优化性能的改进、优化模型的完善以及实际效能的提高，提供了有益的方法和思路，其成果对于群智能计算以及智能计算领域，均具有一定的指导意义与应用价值。