Multi-agent System Based Artificial Immune Network And Its Application

Biological immunity is a parallel, highly evolved, network distributed, self-adapting and self-organizing system, which is provided with special capabilities of learning, recognizing and memorizing, and also provided with a strong mechanism of optimization information that can adapt to a dynamic environment, and the biological immune mechanism is also characterized in that the global and local searching ability are strong. In recent years, the superiorities of the artificial immune algorithm have been shown in solving multi-modal problems, multi-objective problems and other optimization problems. However, with the complexity of immune mechanism, the description of the immune phenomenon is difficult. Therefore, imperfections still exist in modeling and designing of the algorithms. Under the condition of the needs of solving complex optimization problems in the engineering field, agent and multi-agent systems are considered as high-level guidlines for solving complex problems. The combination of agent and immune technologies may provide useful ideas for designing a novel artificial immune network, by using the existing research results of multi-agent to enable the artificial immunity to better adapt to the complex optimization problem.In view of this, this paper has established an immune multi-agent network model on the basis of the biological immune mechanism learning from immune network algorithms and clonal selection algorithms and also the self-governing, self-organization, competition and coordination of the multiple agent system, the paper has also established an immune multi-agent network model starting with the immunity response theory which reflects the biological immune system to better balancing the global and local searching through the clonal selection, competitive collaboration and self-learning operation among the individuals. The paper also discusses the strategies of multi-agent multi-modal global optimization, high-dimensional system optimization and dynamic optimization of immune system by systematic theoretical studies, series of performance tests and comparisions, the algorithm are applied in the modeling of fractionating unit, resources optimization of the fractionating unit and the optimal control of the greenhouse.The paper includes the following four aspects:(1) Study on the Optimization System for Multi-Agent Artificial Immune NetworkThe organism evolutionary system may be taken as a typical complex adapting system, and the biologically inspired optimization searching model should also conform to the complex adapting system model. The complex adapting system model of Genetic Algorithm (GA), Clonal Selection Algorithm (ICS) and Artificial Immune Network Algorithm (Opt-aiNet) typical in artificial immune algorithm is established according to the complex adapting system model proposed by Holland. On this basist the dynamic models of the three algorithms are studied, i.e., the influences of various operators of GA, ICS and Opt-aiNet on mode survival and diversity, thus reveal the roles of various operators in the optimization searching. The multi-modal benchmark verification indicates that ICS and Opt-aiNet algorithms are provided with relatively sound ability of maintaining diversity, particularly favorable for the optimization searching of multi-modal function, especially, Opt-aiNet can better coordinate the balance between global and local searching, thus is regarded as a potential searching algorithm. BDI-reactive immune network optimization searching model AINM-MOD which can adapt to the environmental changes is proposed according to the typical structure of multi-agent and Opt-aiNet, including environment, BDI-inference mechanism and reactive artificial immune network searching mechanism, and the competitiveness and self-confidence of individuals under the environment are defined. The AINM-MOD model uses the operators of the neighborhood clonal selection, neighborhood competition and neighborhood coordination under the grid environment, self-adapting adjustment of self-confidence is possible, the analysis and verifications of the dynamics and diversity of various searching operators of AINM-MOD are performed.(2) Study on the Multi-Agent Multi-Modal Model for Artificial Immune Network and Its ApplicationThe dynamic cloning strategy (DCAS) which may coordinate the global and local optimization is proposed first for solving easy premature of multi-modal optimization, DCAS may automatically adjust the search space and the operating parameters. A multi-scale mutation strategy is used to provide useful ideas for the optimization of multi-modal function. Different cloning strategies are compared for the optimization of the artificial immune network multi-agent model in the multi-modal function, the cloning strategy based on the Sigmoid function is proposed, the Sigmoid function reflects the levels of suppression and expansion between antibodies, and between the antibodies and the antigens, thus reducing the blindness of the optimization searching. Q-mutation strategy based on Gaussian mutation and elitist-learning and?-mutation strategy applicable of improving the global searching capability are proposed. On this basis, the Artificial Immune Network Multi-agent Algorithm(Ma-aiNet) for multi-modal optimization is proposed and convergence of the algorithm is proved in theory. The verification of benchmarks indicates that Ma-aiNet performs better than similar current algorithms in the aspect of optimization accuracy and stability. Study on the optimization of fractionation system resources with the target of minimizing energy consumption has significant influence on chemical production, through analyzing and creating the energy consumption model of fractionation system, the optimization model with the minimum heat load of fractionation system is created to obtain the resource ratio of various fractionation systems, and SQP, GA, ICS, Opt-aiNet, DCAS and Ma-aiNet algorithms are used in optimization of the fractionation system resources. Practical verification indicates that compared with other algorithms, Ma-aiNet performs best in searching and off-line performance.(3) Study on the Artificial Immune Network Multi-Agent High-Dimensional Nonlinear Optimization Model and Its Application In order to solve the problem of the evolutionary algorithm easy falling into the local extremes due to the enlargement of the searching space and strengthening the degree of coupling among the variables in the high-dimensional optimization system, the Artificial Immune Network Multi-agent Algorithm (Maopt-aiNet) is proposed for the optimization of high-dimensional function with double agent network structure, double mutation strategy, dynamic coordination searching strategy and grid self-learning operations. Benchmarks indicate that Maopt-aiNet has stronger searching capability in high-dimensional system. In the application of modeling the fractionating unit, the target function of minimum deviation between the measured temperature of the column plates which represent the separation efficiency of the rectifying tower and the calculation values of the model is proposed. For the evaluation of the time consumption of the mechanism model, the study uses two agent evaluation models, i.e., pure mechanism model and the combination of the neural network and mechanism model to determine the efficiency of the column plate. As shown by practice, under the precondition that the accuracy of the fractionating tower is guaranteed, the searching time of the latter is greatly reduced, therefore the latter is regarded as a promising searching method.(4) Study on the Artificial Immune Network Multi-Agent Dynamic Optimization Model and Its ApplicationIn order to improve the evolutionary algorithm to better adapting the environment changes, the diversity of population will always be maintained in the searching. To this end, two dynamic optimization strategies are proposed, Artificial Immune Evolution Dynamic Optimization strategy based on gradient information (AIDE) and Dynamic Environment Artificial Immune Network Multi-agent Optimization strategy (Dmaopt-aiNet), the former inspied from the immune germinal center-based dynamic response mechanism, which uses multi-population, multi-scale mutation and the operation strategy based on gradient. The strategy of the algorithm also shows superiority in the verification of the benchmarks. The latter strategy adds the environmental detecting factors on the basis of the Artificial Immune Network Multi-agent Algorithm (Ma-aiNet) for multi-modal optimization. Simulation results of the benchmarks show that, Dmaopt-aiNet can accurately search the optimal under the dynamic environment, and has strong capability of evolution and well population diversity. Finally, AIDE and Dmaopt-aiNet are used in the typical dynamic example of the greenhouse control system, and the comparison analysis of the results is made.