The Research On Evaluation Of Digital Human-Machine Interface In Nuclear Power Plant Based On Deep Learning

Safety is the cornerstone of nuclear power development,and the causes of typical accidents are mostly human-factor events since the use of nuclear energy.As an important window for the operators to interact with the nuclear power system,the human-machine interface is playing a vital role in nuclear safety.A good human-machine interface design is playing a vital role in preventing human-factor events.Human-machine interface evaluation,using scientific and reasonable methods to evaluate the qu ality of the interface,can be used as a reference and guidance for the design of the human-machine interface,and effectively analyze the shortcomings of the interface and improve its quality.Therefore,the research of human-computer interface evaluation theory and method has important theoretical and practical significance.In this paper,the non-safety digital human-machine interface of the PWR nuclear power plant was selected as the research object.The necessity of human-machine interface evaluation,the design standards of human-machine interface for nuclear power plants and the existing evaluation system was investigated and analyzed.The human-machine interface evaluation indices suitable for the main control room of PWR nuclear power plant was summ arized.The indices were divided from a multi-finger perspective,and the indices clustering was realized by the hierarchical clustering method,and the human-machine interface evaluation indices system were constructed.Taking the operating process from r eactor power operation to hot shutdown conditions as an example,using traditional evaluation methods and deep learning algorithms to evaluate the human-machine interface.The main contributions were summarized as follows:First,the Gray Relational Analysis was combined with the Analytic Hierarchy Process to verify the indices weight distribution scheme based on the reliability matrix.Aiming at the problem that the reference sequence inherent in this method was difficult to determine,a we ight distribution method based on evaluation indices scores was proposed,which was suitable for the problem of the indices weight distribution of the digital human-machine interface.Secondly,the deep learning algorithm was applied to the evaluation of the digital interface in the main control room of the PWR nuclear power plant.The human-machine interface evaluation model was trained with BP neural network as the core,and the model was optimized by K-fold cross-validation and Genetic Algorithm combined with greedy algorithm.The training has obtained a reliable model,which shows the feasibility of applying the deep learning algorithm to human-machine interface evaluation.Finally,the human-machine interface evaluation system was designed and developed for the main control room of nuclear power plant.By using the system to evaluate the digital human-machine interface,analyze d the evaluation results,optimized the digital control interface,and designed a task-based man-machine interface.The optimized human-machine interface was evaluated by the evaluation system,compared with the interface of the original digital human-machine interface,and analyzed the advantages of the optimized interface.Taking the reactor shutdown operation as an operating example,the correctness of the interface optimization was verified,and then the accuracy of the evaluation system and the rationality of the theoretical method proposed in this paper were verified.