Application of Data-Driven technology in nuclear Engineering: Prediction, classification and design optimization

Currently, workers in nuclear power plants need to monitor plant data in real time. In the event of an emergency, due to human subjectivity, the operator cannot make accurate judgments on the instantaneous state of the nuclear power system based on experience alone, thus missing the best time for emergency repairs, which leads to accidents. This article is based on data-driven technology, for its application in the field of nuclear engineering preprocessing (missing value imputation, image denoising), prediction (critical heat flux prediction, prediction of parameters under LOCA accident, radiation concentration and radiation level prediction of nuclear power plant), classification (troubleshooting, flow pattern identification), design optimization (optimization of fuel loading mode in nuclear reactor core, design optimization of nuclear reactor radiation shielding) four aspects of review. Firstly, this paper finds that some models themselves have certain defects. Therefore, it is an important direction to select a relatively suitable algorithm for different problems in nuclear power plants. Secondly, in the prediction and classification of data-driven technology. Algorithmic models require a large amount of relevant data for training. However, due to the safety problems of the nuclear power plant itself, there are few abnormal data on accidents. Data-driven is a technology based on big data. The lack of real accident data will inevitably lead to inaccurate models. This paper reviews this aspect, and methods such as deep generative networks and separated data sets have good results. Finally, in terms of optimization techniques. Due to the many factors to be considered in the core fuel loading mode and reactor shielding design, conventional manual calculations are time-consuming and laborious. Genetic algorithm and particle swarm optimization are widely used in data-driven technology. Several variants have been developed in recent years. The application of some integrated models and the application of algorithm systems and frameworks for specific objects will have better results than using a single model. It is hoped that the above review content will provide an important reference for the design of nuclear power plants and the engineering application of data-driven technology in nuclear power.