Study On Theory And Applications Of Intelligent Optimization And Immune Network Algorithms In Artificial Immune Systems

The nature immune system, as a parallel, distributed and self-adaptive informationprocessing system with high intelligence performance, provides a new way to dealwithreal time problems. Study on how to perfect artificial immune models reported,establish new ones, and investigate their theory and applications through sufficientlyexcavating, and using abundant resources of the immune system for reference, hasbecome important research contents and main development trend of artificial immunesystems in artificial intelligence. Under this background, a series of novel algorithmsare proposed associated with the problems of optimization, data clustering and signalsimulation through utlizing principles of immunology. These consist of three kinds ofalgorithms, namely, immune algorithms, multiobjective optimization ones and immunenetwork algorithms. They reflect on specifically dynamical characteristics of theimmune system from different angles while enriching and developing the contents ofartificial immune systems. The theoretical study, performance tests, comparisons andpractical applications show that they are not only available but also effective. Theacquired achievements of this dissertation are summarized as follows:A. ImmuneAlgorithmsandMultiobjectiveOptimizationones:TheoryandApplications. A1. A universal immune algorithm is established based on the humoral immuneresponse, while its convergence and converging speed estimation is derived. Throughputting forward the concept of stability for heuristic algorithms with random searching,stability of the algorithm is demonstrated by way of utilizing some properties of intervalfitness functions. Simulation illustrates its robustness and low calculation complexity,while demonstrating its validity by comparisons and applications. A2. Four algorithms with rapidly searching optimal solutions suitable forfunction optimization problems, i.e., niche immune algorithm, dynamical size immuneone, constrained optimization immune one and fuzzy immune control one, areestablished via combining some correlative methods and parts of immune metaphors.The former three algorithms are obtained via combining some simple mechanisms ofclone selection principle into the niche sharing fitness method, the suppression idea ofimmune networks, the golden section method and a method of constrain conditionprocessing, respectively. Meanwhile, the last one is got by introducing fuzzy logicalrules into the second algorithm, which purpose is to improve its searching performance. V英文摘要Convergences of the former three methods are demonstrated theoretically, and theirpractical applications and comparisons made out empirically. At the same time, the lastone is examined its emergence properties through simulation. Besides, a novel neuralnetwork applied to constrained function optimization problems is brought forward thatis compared to the above constrained optimization immune algorithm. A3. On one hand, based on some immune characteristics that antibodypopulations respond to antigen ones, a multiobjective optimization algorithm suitablefor nonconstrained multiobjective optimization problems is established by means ofsome simple mechanisms of the clone selection principle, the idea of immune networksuppression and the concept of Pareto optimality, while its convergence being checked.In addition, a theoretical framework of stability for multiobjective optimizationalgorithms with heuristic random searing is established. Stability of the algorithm isstudied. The theoretical analysis, performance tests and a practical example show that itis both feasible and effective. For the other hand, based on the function of immunememory cells and the interaction relation of antibody populations and antigen ones, analgorithm applied to constrained multiobjective optimization problems is obtained invirtue of the clone selection principle, the regulation mechanism of immune netw...