雨滴计算模型及其应用研究

As we know, it is a key problem in the research of Nature Inspired Computation, that how to set up a self-correcting homeostasis mechanism between the global and the local searching process, and to improve the whole computing ability of finding the optimal solution for complex problems solving with the condition of convergence. Based on the consideration above, this project aims at designing a novel nature inspired model, namely the Raining Computation Model (RCM for the short), for optimization with high computing capacity. Through observing, abstracting and simulating the natural raining process, RCM is inspired by the dynamic process of rain and raindrops getting vertical radial local flow affected by gravity along the terrain, in order to enable its self-adaptation in frequently changing conditions, and to get the ability of surrounding pattern recognition and self-organization under a complex environment. To be more detailed, RCM simulates natural phenomenon such as condensation, diffusion, flow and gather of raindrops, in which the production of new raindrops is based on raindrop's relative distance, and uses a semi-certainty search pattern that raindrops randomly generated make global search with uncertainty and diffusion operation has an effect on neighborhood so that local search direction is determined, meanwhile, the search process for the optimal solution and local extremum value is in parallel computing state, all of which make raindrops full use of implicit information in different time and ensure the convergence and stability of the algorithm. Specifically, at first, the causes of raindrops, makes the raindrops not so frequently fall along the same ground surface on a particular continuous time interval, so that continuously new raindrops realize to the unknown fields search. Secondly, raindrop itself has certain quality, makes it diffusing on the ground surface to realize the local neighborhood search. Thirdly, affected by gravity raindrops constantly flow to the local extremum value to the continuation of the local search process. And finally, raindrops gathered to concave points of ground surface contains the information of all solution for requested complex problems. Thereafter, RCM will be used for the multimodal function optimization, or for multi-object optimizations. Through which, functions on the oriented problem by the corresponding parameters and operators are analyzed, that is also used for its application on guiding the real problem solving. In the end, through obtaining the research results with the authorized properties, some internationally influential academic viewpoints will be proposed in Nature Inspired Computation research area, which will also make the conditions for further work.

在面向复杂问题求解的优化计算模型中如何建立全局与局部搜索之间的动态平衡，提升算法在收敛性前提下的综合计算性能，是自然计算模型研究中的关键问题。有鉴于此，课题组通过观察、抽象，进而模拟自然现象的降雨过程，探讨一种新型、高效的自然计算模型（雨滴计算）的实现方法，以提升计算模型在多变环境条件下的自适应能力，以及解决在相对复杂背景下的应用模式发现和自组织优化等问题。特别是通过对自然降雨过程的抽象，提取人工雨滴模型的特征和作用机理，在刨析全局与局部两种不同性质的搜索过程之间平衡机制的基础上，设计雨滴计算模型的相关规则。在此基础上，进一步通过针对多模函数优化、多目标优化等问题的仿真研究，分析模型中相关参数、算子等对问题求解的作用规律，从而指导算法在实际问题中的应用。最后，通过取得具有我国知识产权的研究成果，在所属研究方向上建立具有一定国际影响力的学术观点，为后续的研究工作创造条件。

本课题旨在探讨如何借鉴自然降雨过程中蕴含的优化机制，构造出一种新型的、高度协同的智能计算模型——人工雨滴算法。具体工作主要从四个方面展开：首先，通过模拟自然现象中的降雨及其雨水在地面局部区域内的动态流动过程，探讨了一种雨滴计算模型的原型实现方法。特别是通过对降雨过程的深入观察和分析，抽象出其自然现象中所蕴涵的内在作用机制，进而提取出雨滴计算的行为特征和作用机理；其次，针对雨滴计算模型中的关键作用机制，即全局优化与局部优化过程的动态、自适应调整，探讨了一种相对简单，但依然能客观反映原始模型作用机理的实现途径和操作方法，从而完成模型设计的关键环节；其三，在上述工作的基础上，分析该系统的收敛性、求解质量、计算复杂度等算法的一些基本特性，进一步构建出雨滴计算模型的理论框架，作为指导后续应用研究的基础；最后，以应用为导向，探究了拓展其集成系统的研究工作，并以此为契机取得多项具有我国自主知识产权的研究成果。.研究工作的主要成果是构造出具有高度协同特性的人工雨滴计算模型。该模型受自然降雨现象的启发，将蕴含在自然降雨过程中的有效信息处理机制移植到优化算法的设计方面，即：在优化过程中，产生水汽或雨滴的位置信息以海拔为评估依据，并在雨滴池中记录海拔较低的位置。同时，引入雨滴池建立存档机制，使产生的雨滴在势能属性的驱动下，自主地向海拔较低的方向流动，从而在整体性能上加快算法的收敛速度；另一方面，在雨滴的碰撞过程中，该模型采用拟正态分布的变异策略有效地维持了种群的多样性。大量的仿真实验结果表明，人工雨滴计算模型在解决复杂优化问题方面能够较好地平衡算法的探索与开发能力。.课题组在研究期间发表论文22篇（其中SCI收录13篇，国际学术会议论文9篇），申请多项国家发明专利，已授权8项，注册软件著作权14项。对照立项时制定的量化目标，已全面保质保量完成。