| The safety of AP1000 passive system has been greatly improved from simplifying devices and reducing dependence on external inputs, instead of relying on natural laws, such as gravity, natural convection and heat transfer, so the uncertain input parameters have more significant influence in the system performance, therefore, physical processes fail become an important factor leading to system operation failure. Meanwhile, the probabilistic safety assessment (PSA) method which so widely used in conventional nuclear power plant safety analysis is no longer fully applicable because of the difference of design ideas and the complex thermal will takes a lot of time to complete the reproducible calculation procedure and conventional neural network method has more stringent requirements to the amount of input network parameters.In view of the above problems, this paper takes the passive containment cooling system (PCCS) in the AP1000 reactor for example. Preliminary screening all parameters that affect system performance so as to get the related elements for the establishment of a hierarchy structure by the way of expert evaluation. Considering the system own characteristics, we improved the traditional analytic hierarchy process (AHP), some parameters which are placed in the bottom in the traditional AHP model are placed in the standard layer in the improved AHP model, making it more suitable for the evaluation of the impact of natural recycle system. In addition, the paper will establish the hierarchy structure according to the traditional AHP model with the above-mentioned preliminary screened parameters and compared with the improved AHP model, the results show that although the weight of all parameters have slight variations, but the key parameters which have the greatest influence in the target layer remain the same, namely break diameter, steam temperature, atmospheric temperature and steam flow.According to the distribution of the input parameters, the author random sampling each key parameter to obtain several sets of input data and get the output combinations of corresponding key sensitive parameters with the thermal program, and obtain representative training sample sets of artificial neural networks based on orthogonal design method. Finally achieved the alternative thermal computing model by training the important parameters of neurons in neural network nodes, improve the computational efficiency.Moreover, considering the important influence of atmospheric temperature on the AP1000 PCCS, and its probability density curve is derived based on historical data now, while the emergence of extreme heat has an extremely low probability, and exists greater impact on the evaluation results, the paper analyzes the meteorological data based on the cluster classification algorithm from writing Java program, establish association rules, acquire the law of high temperatures appear, and laid a good foundation for achieving the temperature probability density curve with the system evaluation method. |
