Dynamic Decision Modeling Theory And Applications

Decision modeling and solving is currently a research hotspot of the artificial intelligence field, and has become one of the core technical support of robotics and expert systems. The purpose of decision analysis is to provide a reasonable, intelligent reasoning process for complex decision-making problems with uncertain factors, assiting decision makers to select rational behavior in an uncertain environment. In particular, the probabilistic graphical model theory is the main branch of the field of decision analysis, which uses a clear graphical representation to model and solve complicated decision problems. Decision-modeling techinology provides an effective decision analysis modeling tool, advancing the development of Decision-making theory.In order to solve decision problems with time constrains and multiple agents, this thesis propose a new probabilistic graphical model theory: Knowledge-based time critical interactive dynamic influence diagram (I-TCDID) modeling theory. Combined with the techonology of I-DID and TCDID, I-TCDID introduces the concept of model-abstract and time-abstract. Meanwhile, based on the theory of Knowledge-Base and Meta-Resoning, this thesis proposes inference algorithms and model-selection algorithms to achieve the best trad-off between decision qualities and compute complexity. Also, we introduce machine learning algorithms according to different decision-making quality and time requirements, achieving different levels of the model solution. In addition, this theory is applied to decision control in the anti-air defense domain, verifying the correctness of the proposed theory, and boosting the development of the anti-air defense system. This thesis immerges further into the theory of multi-agent and time-constrained decision-making theory, and broadens the scope of its application areas, theoretically breaking through the probabilistic graphical modeling technology.