参数扰动或概率约束优化问题的微种群免疫协同进化算法及应用研究

Uncertain programming as a challenging scientifical problem has increasingly presented its prominent importance in the field of optimization and become another goal to be attacked in the research branch of intelligent optimization. When solving dynamic porgramming problems, immune optimization theory and applications under uncertain environments as an important exploring reseach branch of artificial immune systems has exhibited its unique superiority and potential motivation. However, less research work on constrained programming under parameter perturbation or probabilistic constraints is done rarely from the angle of immune optimization. Thus, based on the immune theory which has attracted much attention among researchers, this proposal will probe into micro immune coevolutionary optimization approaches and their performance analysis, convergence and applications for general single and multi-objective nonlinear optimization problems of interval number programming and probabilistic constrained programming, relying upon developing new immune coevolutionary models related to biological ideas of immune response. Additionally, it will also study applications of such approaches and effectiveness for fly visual neural network based collision detection models acquired, and make such models become simple and valuable. Originally theoretical achievements help for speeding up the development of artificial immune systems and motivating researchers to track the development trend of immune optimization. Applicable achievements can play an important role in providing new collision alarm techniques for vehicle. The achievemnets to be acquired are very meaningful for solving practical problems and activating the development of correlative interdisciplinary.

不确定规划作为最优化领域极具挑战性的科学问题已日显其突出地位，并已成为智能优化研究分支攻克的又一亮点。以生物免疫理论为依托的不确定环境下免疫优化算法理论与应用作为人工免疫系统衍生的又一探索性研究分支，在解决动态规划中已展现其独特优越性和潜在活力，但在参数扰动或概率约束优化问题方面的研究尚属罕见。为此，本项目拟基于倍受关注的免疫学理论，针对一般单、多目标非线性区间数规划及概率约束规划，提取免疫应答思想，建立新免疫协同进化模型，展开微免疫协同进化算法的设计、收敛性、性能测试及应用研究。探讨算法在已获果蝇视觉神经网络碰撞检测模型中的应用，优化该模型的结构，验证其在碰撞检测中应用的有效性。源头性理论成果对人工免疫系统自身发展具有重要推动作用，有助于激励国内外智能学者追踪该方向的研究动态；应用成果有助于为碰撞预警（车辆）提供新技术。研究成果对实际问题的解决及相关交叉学科的发展具有重要现实意义。

针对具有广泛工程应用背景的不确环境下参数扰动或概率约束规划问题，基于区间分析的基本理论，探究并获得区间数、区间向量值空间的代数性质；借此，给出区间数规划解的概念及研究此类问题与相应自然区间扩张规划问题的内在联系，获得一条可通过自然区间扩张规划求解区间数规划的快捷、有效新途径，突破了长期仅能利用嵌套式优化结构或模型近似化方法处理单、多目标区间数规划的研究思维的束缚。其次，借助小种群进化、协同进化及免疫学中免疫应答原理或机理，针对一般类型单、多目标约束或非约束非线性区间数规划、区间值规划问题，从源头上系统地探讨求解的微种群免疫协同进化模型和算法设计，展开算法的理论研究、性能测试、工程应用等研究，已获应用性强的高性能源头性算法成果，此对推动免疫优化及进化计算的发展将发挥重要的助推作用。再次，针对实时环境下车辆智能预警与异常检测问题，探索并获得刻画碰撞、异常行为的视觉神经网络，其参数设置由免疫优化算法确定，由此获得智能预警样机，并已申请技术发明专利。在协作单位支持下，针对该单位的石油计量问题，利用免疫优化成果优化计量模型的参数设置，所获成果已在油田行业投入应用并已以技术发明专利呈现。该项目已取得多项原创性研究成果，部分成果已跻身国际前沿之列；研究工作和成果正受到同行研究者的跟踪和认可，为免疫优化的纵深发展提供了发展动向、新研究方向和新应用领域，也必将为人工免疫系统、交叉学科的发展和工程问题的解决发挥积极作用。已发表标注本项目基金号的学术论文33篇，其中，SCI检索的一区论文1篇，二区论文4篇，三区论文2篇，四区4篇，EI检索杂志论文5篇，中文核心刊3篇、国际会议4篇（其中EI收录2篇），普刊10篇，待EI收录杂志论文1篇；培养已毕业博士4人及硕士13人，青年教师2人。已获技术发明专利1项，申请并已受理技术发明专利1项。