Research On The Bayesian Networks Methods For Situation Assessment

Situation Assessment (SA) is an active research field with great challenge which has evolving over twenty years. In the process of SA development, any new headway of artificial intelligence theories and techniques has promoted SA researches and made great progress. However, it's greatly constrained because of the difficulties existing in knowledge representation and reasoning algorithms and there hasn't a general theory framework. With the development of the Bayesian networks (BNs) on knowledge representation and reasoning algorithms, the methods for SA based on the BNs technology become a hot topic in the domain of SA. This dissertation researches on the main questions for the methods for SA based on the BNs technology and constructs military SA system based on BNs. The main points can be summarized as follows.1. Researches on the functional model and analyzes the events and groups to process in each stage of SA. For the shortage that the situation hypothesis models can't represent the opposite activities of ownship, presents an interactional situation hypothesis model, indicates that SA is a process that confronting and interacting between the enemy and ownship. On the basis of the representation of BNs models, presents the mathematic model and its description of SABN, expatiates the connotation and elements of the model, analyzes the character of the model, indicates that causal BNs are more appropriate for the SA process.2. Researches on the three main processing methods for uncertain information, i.e. the Subjective Bayes Methods, the Fuzzy Sets Theory and the D-S Evidence Theory. Analyzes the processing methods for uncertain information and represents these three uncertainty reasoning methods in a uniform framework with the BNs. For these three methods, presents the representing methods in the BNs for the uncertainty of the reasoning rules and the methods to calculate the conditional probabilities. Finally, presents the representing methods in the BNs for the evidence' uncertainty obtained by the Fuzzy Sets and the D-S Evidence Theory.3. Analyzes the difficulties existing in the learning methods of BNs for SA. For the existing questions, presents a method based on the combining of TOP-DOWN and BOTTOM-UP methods to construct SABN automatically. Presents the hierarchical and layered SABN models, defines the BNs' modules which comprise the model and classifies the modules, discusses the structure of the templates library of the BNs' modules and researches on the constructing methods for different types of modules. For the set of military events acquired in the process of SA, presents searching and instantiating algorithms for the BNs' modules and the combining algorithm after instantiating to construct the SABN. On the basis of these contents, presents an illustrative example to verify the validity of the methods.4. Analyzes the existing temporal reasoning methods, indicates that the temporal reasoning methods based on the Temporal Interval Theory are more appropriate for SA. For the difficulties caused by the complexity of the network structure and the reasoning methods, rebuilds the Probablistic Temporal Theory (PTN) based on the basis of the Temporal Interval Theory and BNs technology. Inducts temporal semantic to rebuild the BNs as the Temporal Bayesian Networks (TBNs), constructs the definition system for TBNs and shows the methods to rebuild the traditional BNs applying the TBNs by an example. Presents the model and definition for the mutual exclusive processes which are the most common in the process of SA and applies the TBNs for temporal reasoning.5. Designs and realizes the STAB system for SA with the methods presented in this dissertation. The system is generic and expansible as it is not depend on the specific scenario.