Theory And Application Of Immune Optimization Algorithms In A Random Environment

Studies on intelligent optimization for solving static optimization problems has made great progress,by which some important and mature schemes can be adopted to handle complex engineering problems.Further,dynamic optimization as a hot topic has gained enough attention among researches in recent years,due to a large number of innovative achievements.Stochastic programming,which differs from static and dynamic optimization,is a special type of uncertain programming with noises.Since stochastic factors seriously disturb the process of solution search,it has not gotten great progress,in particular its algorithmic research.Therefore,this dissertation investigates several general kinds of stochastic programming models,including single-objective nonlinear expected value programming without or with constraints,nonlinear multi-objecive expected value programming without constraints and nonlinear single-or multi-objective expected value programming.The main work concentrates on their lower bound estimates models for samples,adaptive sampling algorithms and quality discrimination models for individuals,while probing into their intelligent optimization algorithms from the angle of immune optimization in artificial immune systems.These algorithms are also comprehensively discussed,involving their computational complexity,comparative analysis,parameter sensitivity,algorithmic applications and so forth.The acquired achievements can give researches some new insights,also enrich and develop connotation-based artificial immune systems.They can also provide some potential and valuable optimization techniques for engineers to solve engineering optimization designs.The main achievements can be summarized below.A.The problem of single-objective,multi-modal expected value programming is transformed into a special multi-objective expected value programming problem by introducing a stochastic performance index which examines local optimal solutions in stochastic environments.An efficient path is acquired to seek multiple local or global optimal solutions after the relation between solutions for the two problems is studied.On the other hand,a new multi-objective immune optimization algorithm,with the merits of effective individual comparison and strong noise suppression,is developed to solve the sample-dependent approximation model of the multi-objective problem by designing a recurrent non-dominated sorting method and an adaptive sampling model,depending on the mechanism of immune learning involved in the principle of clonal selection.Its performance analysis is done including computational complexity,performance examination and engineering applications.Numerical experiments have validated that it is available to solve the single-objective optimization problem by transforming it into a multi-objective optimization problem,while the algorithm can find all the local and global optimal solutions with a high probability.B.After the relation between solutions for the problem of single-objective expected value programming and the corresponding sample-dependent approximation model is discussed,two lower bound estimates models are derived,in order to design the two related algorithms used in estimating the objective expected values and constraint estimates of individuals.For the problem of single-objective chance constrained programming,the objective expected values of the individuals are estimated by means of an optimal computing budget allocation approach,while an adaptive sampling detection algorithm is designed to detect individual's empirical feasibility.After that,two new adaptive sampling immune optimization algorithms are developed to respectively solve such two single-objective expected value programming problems,by the mentioned objective and constraint handling methods and also the metaphors of the clonal selection principle.Some theoretical and experimental analyses are done,including computational complexity,comparative numerical experiments,applications and so forth.Numerical experiments have showed that,when solving the optimization problems mentioned above,these two optimization algorithms have significant advantages with the aspects of noise suppression,search effect,efficiency and potential application value.C.A lower bound estimate model is acquired by means of the relation between the problem of unconstrained multi-objective expected programming and the related sample-dependent approximation model.With the help of such model,an adaptive racing ranking approach is designed to pick up excellent and diversity elements from a given population,relying upon a new polymerization degree model.These designs,as related to the interactive immune metaphors between B cells and T cells in the immune system,are utilized to develop a new multi-objective immune optimization algorithm for the above multi-objective expected value programming problem.Some studies on one such approach are done,e.g.,computational complexity,performance examination,applications and so on.Numerical experiments have indicated that such algorithm has the potential value and significant advantage for solving complex multi-objective expected value programming problems.D.After the relationship of solutions for the problem of multi-objective chance constrained programming and the related approximation model is intensively discussed,a lower bound estimate model is developed to control the size of sampling for stochastic variables.Correspondingly,a Bernstein adaptive racing ranking approach is designed to deal with chance constraints.Subsequently,a new multi-objective immune optimization algorithm is proposed to solve such expected value programming problem,by means of some immune metaphors involved in the danger theory in immunology.Its performance characteristics,e.g.,computational complexity,comparative analysis and applications,are studied.Numerical experiments have showed that the proposed algorithm can effectively cope with benchmark problems and has the potential for handling engineering problems.