Study On Models Of Multi-criteria Group Decision-making Towards Formal Safety Assessment

The potential of Formal Safety Assessment (FSA) in International Maritime Organization (IMO) rulemaking process has been gradually acknowledged, and China lists FSA as one of the IMO hot issues worth long-term tracing and researching. This dissertation studies some problems confronting FSA development and finds certain features of FSA study, such as high dependence on experts, imperfect historical data, incomplete model, uncertainty in evaluation methods, inadequacy of evaluation bases, and discrepancies to be eliminated in process, etc. The focus is on the studies of the measurements exploited by experts in evaluating quality, of human errors, of polymerization of experts’semantic evaluation, and of selection methodologies concerning multi-index risk control options (RCOs). Ideas, methodologies, and models of solving the problems are also proposed.Concerning expert evaluation measurements, it is proposed that measures should be conducted in respects of repeatability and accuracy owing to the subjectivity, obscurity, and uncertainty of expert evaluation. The dissertation measures the uncertainty of expert evaluation employing transference entropy, and proposes a method of improving degree of evaluation agreement. It also obtains experts’weights using transference entropy model and Euclidean distance matrix model, aggregates experts’ weights with the construction of dispersion functions, and establishes well-targeted models of expert evaluation assessment and objective weight obtaining.With regard to the problem of lacking historical data of human errors and effective models for quantitative analysis, the dissertation obtains frequency of pertinent human errors using expert questionnaires. Experts’semantic evaluation is converted to trapezoidal fuzzy numbers to determine the probability of occurrence of human errors. The dissertation clusters experts, using a global quick K-means clustering algorithm, in order to demonstrate the feature that subjective evaluation is influenced by experts’ personal background and qualifications. Experts’integral weights in terms of clusters are determined with Euclidean distance matrix model, according to the theory of relations between reproducibility and accuracy, based on every cluster center, and with reference to the optimal expert. Owing to the obscurity of expert evaluation, the obscure weights of in-cluster experts’semantic evaluation are determined with fuzzy similarity analysis model to aggregate the results of in-cluster experts’semantic evaluation. On the above basis, in-cluster aggregation followed by inter-cluster aggregation of experts’ semantic evaluation is executed to obtain the ultimate trapezoidal fuzzy numbers of experts’semantic evaluation. A systematic expert weight obtaining model is established based on clustering algorithms and fuzzy similarity models, increasing the objectivity of expert weight obtaining. Meanwhile, empirical formulas are employed to determine the probability of occurrence of human errors.A selection model of RCO based on dominance degree is proposed. Experts’ understanding of the importance of RCO selection indexes is acquired through FSA experts’ questionnaires to represent decision makers’subjective decision orientation in the process of maritime security policy institution, with reference to analytic hierarchy process (AHP). Weights of every index are gained with certain algorithms, and the dominance degree of each RCO is calculated via the model on this basis. The model not only makes full use of samples’order relationship on each index, but also takes into consideration the degree of difference, thus overcomes the disadvantage of strong subjective factors in the process of RCO selection in FSA studies. The model’s calculation results are direct and clear, facilitating decision making. It is found that GCAF as the preferred reference index has apparent screening effect in dual-index RCO ranking, but unapparent screening effect in multi-index RCO ranking which requires the model to improve screening effect.