Intelligent Optimization Based-on Graphical Models

Graphical models are increasingly popular tools for modeling problems involving uncertainty. In this dissertation, they are integrated with evolutionary mechanism, to develop the evolutionary mechanism based-on graph models and more intelligent optimization theory. This dissertation characterizes promising solutions of population from simple to complex structure by means of grading graphical models. The metric rule of local graphical models is also analyzed so that the technique of speedily and effectively to create a graph is derived. Finally the optimization of hierarchical functions and the path planning for UCAVs(Uninhabited Combat Air Vehicles) are studied by the optimization mechanism that is presented in this dissertation.The main contributions of this thesis are as follows.(1) From the point of view of graphical models, compact genetic algorithms use the simplest graphical structure to model promising solutions. So the genetic evolutionary mechanism on compact genetic algorithms is firstly studied, the evolutionary intension on algorithmic is analyzed, and the corresponding advanced algorithm is put forward. Since to generate positive or negative evolutionary individual pair without intension coefficient, it is showed that the effective evolutionary individual pair of the compact genetic algorithm can decrease. To increase the intension coefficient, and speed up the significant evolutionary , the increases probability of arriving vector optimization is given.(2) It is described how to create the tree models by using information entropy and evolutionary algorithms based-on the mutual information tree. The drawback of those algorithms is presented, i.e. the nodes of these graph models have no obvious causality. The local structure of chain-type models and tree models based-on Bayesian Dirichlet metric are mainly studied, and the theoretic foundation for creating those models is provided. Simultaneously, the algorithms to create chain-type and tree models are presented. These chain-type and tree models to be investigated have obvious causality with high posterior probabilities.(3) The inhere drawback of traditional algorithms to construct Bayesian networks is pointed out. So the local Bayesian network metric attribute based-on Bayesian Dirichlet metric is also researched, and some notions on Bayesian networks arepresented, such as comparability, independency and symmetry on nodes etc. It is showed that the comparability is an important factor on whether two nodes is linked, and a sufficient condition for nodes with most comparability is obtained. By using this analysis on local Bayesian network, the important principia are presented to create graph models.(4) The design on hierarchical optimization functions is studied, and the fraudulence of hierarchical decomposable functions to algorithms is mostly discussed. So the test tools for optimization theory are presented. The results of simulation show that the hierarchical optimization functions have stronger deceptive so that the algorithms be pendulous among local optimizations. However the evolutionary mechanism based-on graph models being discussed displays its favorable characteristic of intelligent optimizing, such as to overcome deceptive and explore inherent laws on search space.(5) A method for designing a model framework of situation awareness for UCAYs based on object-oriented Bayesian networks is presented. This method is more effectively to explore situation awareness model for multiple UCAVs cooperative combat, and to explore the situation awareness of the air combat environment where there are many threatening entities.(6) Since the same threats are only took into account in the path planning for UCAVs based-on Voronoi graph, this dissertation develops a path planning scheme based-on different threats for UCAVs. It establishes the principle which can selects the path in local area having different threats, and presents a new approach to create the path diagram.