Research On Evolutionary Decision Making: Its Models, Key Techniques And Applications

Decision Making (DM) is defined as the purposive act of choosing one from multiple alternatives, such as judging, choosing, and determining. Decision making is one of the most elemental behaviors of human being. General DM techniques often play supportive roles in engineering fields as information, control, economy, energy, transportation, manufacture, and society management. The performance of decision making, which is critical to the success of scientific, engineering, and social practice, is greatly dependent on the effectiveness of the methods used. Modern DM techniques have developed into a comprehensive integration of the theories and methods of System Modeling, Statistics Analysis, Optimization Theory, and Computer Science, etc. It is obvious that researching on DM techniques is of great significance.With the popularity of computer and computation methods, more and more complicated computation and reasoning are relied on computers. Consequently, the Intelligent Decision Making (IDM) problem is put forward in information processing, robot control, and experimental science, etc. That is, how the intelligent agent can make decision automatically to complete its mission.Almost every intelligent behavior is based on decision making, making which becomes the fundamental problem of human intelligent. How to develop specific-purpose and/or general-purpose decision making methods are an elemental challenge faced by Artificial Intelligence (AI) researchers. The purpose of IDM research is not only to make decisions automatically in place of human being, but also to make decisions under complicated situations where it is difficult to make decisions even for human experts.Recent years, some new kinds of decision making problems are proposed in intelligent control, business, finance, and experimental science research, etc., which have the common characteristics of huge body of data, uncertain environment, requirement of adaptation to changing environment, automatically decision making, real time constraints, and high reliability requirement, etc. Those new problems bring great challenges to current IDM techniques.Being a new kind of general problem-solving paradigm inspired by the principle of biology evolution, Evolutionary Computation (EC) is meanly used for solving optimization problem, and has found great many of successful applications in Machine Learning (ML). Though its remarkable success, its potential for solving complicated practical problems are still largely undeveloped. Since IDM is based on the constructing and computing of computable decision rules, for any IDM problem, if the general mathematical model of its decision rules can be constructed, then solving of the intended IDM problem is reduced to searching the best parameters for that model. Thus EC can be made used of.The concept of Evolutionary Decision Making (EDM) is proposed in this thesis. Which refers to the theory, methods and applications of using EC, combining with Decision Making Analysis (DMA), to solve IDM problems. The key point of EDM lies in EC's capacity of automatically modeling while the formal computable models are used as materials of evolutionary learning. EDM has some remarkable advantages over traditional models, includes using implicit causal models, self-learning capacity, weak dependence on domain knowledge, wide applicability, robustness, self-adaptability, and population-based searching, etc.Tracing back its intrinsical ideas, EDM is just making use of the nature's decision making strategy, natural selection, to solve the decision making problems faced by human or the intelligent agents. Since evolution is by now the most powerful problem-solving paradigm we knew, the prospect of the development and application of EDM in the future is expectable.Two fundamental problems are presented by the application of EDM. One is how to design the intended computable models of intelligent decision making rules that are suitable for machine learning, for the decision making problems considered. The other is how to...