Immune Optimization Algorithms Based On Memory-Evaluation-Guide Mechanism

Abstract:An artificial immune system mimics the structure of a biological immune system and it is a type of the artificial intelligence system with high-performance, self-organization function, and strong robustness. Optimization problems widely exist in the engineering practice and the theoretical research. During the treatment of the optimization problems, most intelligent optimization methods pay less attention on the memory and usage of the non-genetic information. Based on biological principles of the immune memory and the clone selection, this thesis designs a memory-evaluation-guide mechanism, which concerns on the collection and usage of the non-genetic information, and the thesis builds an immune optimization algorithm model based on the proposed memory-evaluation-guide mechanism. Based on the proposed model, several different artificial immune algorithms are built for the different types of optimization problems.The major work and the innovative achievements can be divided into the following four aspects:Firstly, an immune optimization algorithm model based on memory-evaluation-guide mechanism is constructed. The proposed novel model is constructed by combining the principle of biological immune and the framework of the existing artificial immune algorithm. The model attempts to combine the processes of the evolution and the learning, namely, it considers not only the Darwinian evolution but also the effect of non-Darwin with the characteristic of the experiential learning and the experience genetic. Compared with the traditional optimization model, the presented model collects and uses the information of the exploration, which greatly reduces the repeated search and blind search and then improves the convergent performance.Secondly, a single-objective immune optimization algorithm based on the accurate information memory, so-called the mutation memory matrix based clonal selection algorithm (CSAB3M), is developed. The proposed algorithm applies the immune memory matrix to save the useful mutation information, which can be used to guide the cloned offspring and mutation operation, thus the local search ability is strengthened. The global search ability is improved by using the comprehensive information of the contemporary population to produce a new generation of antibodies, and the accuracy of the algorithm is improved due to the self-learning of the optimal antibody. Simulation results for the standard test functions show that the algorithm is suitable for solving the complex function optimization, and it has many advantages such as the rapid convergence, the powerful global search ability, high accuracy, and good robustness.Thirdly, a new type of the single-objective immune optimization algorithm based on fuzzy information memory is presented, including an immune algorithm based on adaptive selecting dimension (IABASD) and a clonal selection algorithm based on grade variation (CSABGV). The former divides the mutation scale into several levels and set a matrix to record the information of mutation ranks, mutation results, mutation times etc., and the information is used to guide the selection of the mutation dimension and the generation of the mutation scale for the next generations, which can improve the overall performance of the algorithm. The later algorithm increases the complementary mechanism on the basis of the former and simplifies the procedure of mutation. The simulation results for the broader test functions shows that the proposed algorithm can provide better global convergence performance in comparison with other intelligent algorithms.In addition, nine functions in different dimensions are tested by using two kinds of memory techniques, including the fuzzy information based memory and the accurate information based memory. The results show that the immune optimization algorithm based on the fuzzy information memory has slight advantage over the other one when dealing with the testing problem for the complex and high-dimensional functions.Finally, a multi-objective immune algorithm based on non-Darwin effect (NDIA) is proposed. For the multi-objective optimization problems, in order to strengthen the heritage and usage of the non-genetic information, the range memory mutation matrices are used to save the information of the successful mutation range and then guide subsequent evolution operations, which can strengthen the ability of local search. The algorithm uses Pareto order to choose non-inferior solutions, when the number of non-inferior solutions bigger than the preset value, the order constructed based on crowded distance are applied to choose the relatively sparse antibody. The algorithm defines a homogeneous degree enhancement operator to reduce the final non-inferior solutions, and the homogeneous degree in the target space is increased after the repeated deletion of the most crowded antibodies. Based on the results of the simulation tests on several multi-objective optimization problems and the comparison with many classical methods, the resulting set of solutions provided by the proposed method are greatly improvement in the spread ability, convergence, and diversity and they can converge to global Pareto optimal front quickly. There are35figures,25tables, and131references.