Application Of Artificial Neural Network And Immune-Genetic Algorithm With Elitist Model To Pursuit Problem Of Multi-robots System

The mechanism of the multi-agent system's control has been aninteresting research project in the fields of robotics and artificialintelligence. It incorporates many important subjects, such as intelligentcontrol, artificial life, evolutionary computation, robot programmingtechnology, etc. With the development of research on multi-agent system,it has become a trend to conduct the optimization design and performancesimulation of multi-agent system research based on the artificial life aswell as evolutionary algorithms.In the past research on multi-agent system there are severalshortages. The first is that the existed performance criterions of the preyencircled are too simple to obtain convincing results of performancesimulation by these criterions. The second is that the robots were lackingin enough intelligence and strong learning skills by themselves.Aimed at these shortages, we have developed an innovative behaviordecision-making system, called the IGAE-ANN, based on the evolvableartificial neural network (ANN) and the artificial immune algorithm withelitism (IGAE) for the control of the multi-agent system. Using thissystem we executed performance simulation and parametric design of themulti-agent pursuit problem.In this thesis, first the research situation and common dynamicresearch methods of the multi-agent system are introduced. Then thebasic structure of the system are also designed and expatiated. Thissystem used a feed-forward neural network with three layers to processthe perceptive information of every predator agent and made a decisionfor its action. The IGAE was used to optimize the connection weightvalues of this neural network, which made the performance of the neuralnetwork to be evolved continuously and finally a behaviordecision-making system with good performance could be obtained. Theresults of simulation experiments indicate that the mobile predators cankeep stable encircling of the prey, which proves the gooddecision-making ability of the IGAE-ANN behavior decision-making system during the pursuit game. By comparing with the behaviordecision-making system based on the Canonical Genetic Algorithm withElitism (CGAE), it has been confirmed that the proposed method hasbetter performance in convergence speed, solution variation, and dynamicconvergence behavior.A new criterion of the pursuit problem in the multi-agent system ispresented to make sure that the prey is in the encirclement that can not begot away from. The simulation results show that the presented criterion ismuch more persuasive than that proposed by Korean scholar Malrey Leein the experiment of the stable encirclement of the prey.