A Study on Human Reliability Dynamic Evaluation Model for Human-Computer Interaction in a Control Room of Marine Floating Nuclear Power Plants

The environment for human-computer interactions in the control room of a marine floating nuclear power plant (MFNPP) is more complex than that of a land-based nuclear power plant, as the reliability of the human-computer interaction directly affects the occurrence of human error accidents. Currently, some methods for evaluating human reliability in human-computer interactions are traditional; however, they fail to adequately reflect dynamic and intelligent characteristics.To address this issue, this paper constructs a model for evaluating the human reliability in human-computer interactions in the control room of MFNPPs based on the backpropagation neural network theory, which provides a dynamic method for assessing the human reliability in human-computer interactions and resolves the limitations of static methods. This study first establishes the influencing factors by reviewing the literature, conducting questionnaires, calculating relative importance, and consulting experts, then determines the network structure and calculation methods. Later, the expected values are obtained through the linear weighting method. Two different weights of the same influencing factors were initially determined using the analytic hierarchy process and the decision experiment and evaluation experiment method, respectively. Then, the game theory combination weighting method is used to synthesize two different weights to determine the final weight, with the weight results reflecting the importance and correlation among the influencing factors. Finally, the model is trained. From the perspective of model performance, the constructed neural network evaluation model has good accuracy and can be used to evaluate human reliability in human-computer interactions.