Research On AIS Algorithm Based On Biological Immune Metaphor

Artificial immune system (AIS) is an intelligent method of simulating the naturalimmune system, which realizes a natural defense mechanism inspired from biologicalimmune system, and provides noise tolerance, teacherless learning, self-organizing,memory and other characteristics. Moreover, equipped with advantages as neuralnetwork, classifier, and inference system, AIS provides a novel solution to realisticproblems. The involved research achievements include optimization, control, dataprocessing and fault diagnosis etc. AIS has been a new hot topic in the field of artificialintelligence since neural network, fuzzy logic and evolutionary computation.However, researches on the model of AIS are quite few since the biologicalimmune system is very complex. Although the existing researches demonstrate the greatpotential of artificial immune algorithm in solving some practical problems and theadvantages in handling some optimization problems, current research on AIS is still inits infancy. Its broad application prospect still need more detailed development. In thefield of optimization, researches on AIS mainly focus on improving existing algorithmsthrough realizing immune mechanisms. Although most of these algorithms are titledwith "immune", they are essentially just the modified versions to the genetic algorithmbased on immune mechanisms, and most of them are static and non-adaptive. Moreover,comprehensive analysis on biological immune mechanism and exhaustive comparisonwith other natural heuristic algorithms are also ignored. Based on the existing clonalselection algorithm, learning recognition and defense mechanism of biological immunesystem are elaborated in the thesis. Through extracting the relative immune metaphor,the AIS-based algorithms are proposed to solve the constrained optimization anddynamic optimization problems. The main achievements and innovation are as follows:①Based on immune metaphor, global optimization, constrained optimization andoptimization in dynamic environment are mapped with biological immune responsefrom different angles. A novel biological perspective and solid biological backgroundare provided to solve these problems. Stable information processing and reliabledefense mechanism are elaborated. We construct the AIS based algorithm on account ofcarefully exploring the functions and mechanisms of biological immune system ratherthan borrowing the "immune" concept directly.②An immune-inspired algorithm based on information transfer (IAIS) is proposed. Immune metaphor is extracted from the dual roles B cells played in innateimmunity and adaptive immunity. Accordingly, the analogy between the mechanism ofbiological immune response and constrained optimization formulation is drawn.Individuals in population are classified into feasible and infeasible groups according totheir constraint violations that closely match with the two states, inactivated andactivated, of B cells in the immune response. Feasible group focuses on exploitation inthe feasible region, while infeasible group facilitates exploration along the feasibilityboundary. Although adopts only the traditional artificial immune operator, the IAISalgorithm with proposed framework and information transfer strategy shows greatpotential in solving constrained optimization problems.③The IAIS algorithm was modified to solve the constrained optimizationproblems. On premise of great potential of immune-based algorithm in handlingconstrained optimization problems, some operators of IAIS are modified and theperformance of the modified IAIS algorithm is further improved. In order to overcomethe disadvantages that traditional clonal selection algorithm with prematureconvergence and limited search precision, Recombination operator and Recruitmentoperator are introduced, and hypermutation operator is modified. Meanwhile, Directioninformation extraction is adjusted. The modified IAIS is validated to be competitive incomparing with the state-of-the-art algorithms. The proposed algorithm compensatesthe weakness of the existing artificial immune algorithms in solving constrainedoptimization problems, and promotes the development of artificial immune algorithmitself. Through a lot of statistical experiments, the Modified IAIS is verified to haveglobal searching ability, high accuracy and good stability.④To solve the optimization problems in dynamic environment, animmune-inspired algorithm based on quasi-gradient, clustering and memory mechanism(GCMAIS) is developed. On the basis of careful analysis, shortages of the generalframework of AIS in solving dynamic optimization problems are summarized at first,and then three coping mechanisms from different aspects to improve the performance ofalgorithm are put forward. Firstly, in order to improve the search speed, quasi-gradientinformation is extracted from the redundant information carried by clones. TraditionalJacobian vector and the Tangent vector are extended through the quasi-gradient toextend the search performance. Secondly, in order to improve the search ability of thealgorithm and maintain the diversity of population, clustering method is adopted formulti-population mechanism. The interaction among individuals in each subpopulation and among subpopulations is enhanced. The multi-population mechanism not onlydecreases the redundant information of population, but also promotes the precise search.Thirdly, in order to deal with dynamic characteristics of the cyclic and noncyclic changein dynamic environment, the immune memory mechanism is elaborated. The long-termand short-term memory mechanism is established according to the different life span ofmemory cells. Short-term memory extracts important information from last scenario,which is benefit for tracking the less dramatic changes, while long-term memoryretrieves history information from previous environments, providing references forpopulation initialization under cyclic environment or with recurrent change.Experiments validate the efficiency of the proposed method.