Data Driven Methods for Break Size and Location Estimation in LOCA Based on Deep Learning

Abstract

Abstract Loss of coolant accident (LOCA) is one of the most important accidents in thermal hydraulic design of nuclear power plant. The traditional analysis method is difficult to realize the rapid prediction of the size and location of the break in the LOCA, while machine learning provides an idea for the rapid diagnosis of the initial cause of LOCA. In this paper HPR1000 which is independently designed by China was taken as the object, the process of LOCA under various break location and sizes is studied by using the advanced reactor system analysis code ARSAC (Advanced Reactor System Analysis Code). The key thermal hydraulic parameters including temperature, pressure, flow rate and water level in the transient process of LOCA are analyzed, and a series of data sets are generated. Based on convolution neural network algorithm of deep learning, the transient thermal hydraulic parameters of HPR1000 under different working conditions are learned and the accident diagnosis model is obtained. By comparing with the test data, the trained model can quickly and accurately predict the size and location of LOCA, which can be subsequently extended to the diagnosis of other kinds of accidents in nuclear power plants and has certain application prospects.