Delayed-feedback artificial neural network applied in multi-parameters thermal-hydraulic system performance assessing: An integral PCCS test case-study

Abstract

For validating the safety of the new design of large passive nuclear plant, a series of thermal-hydraulic(T-H) test facilities were set up by SNPTRD. The T-H performance of passive containment cooling system was deeply researched by the sub-scaled modeling integral test system – CERT, according to scaling-analysis method. Besides of traditional T-H methods, we had tried applying ANN to simulate the performance of CERT, which is a multi-parameters system. The ANN is adapted from a Feed-Forward Back-Propagation network with delayed output feed-back. Two different networks were set up for the steady-state and transient case respectively. A directly Monte Carlo(M.C.) selecting procedure was generated to achieve a set of optimized weights of the networks. After the circulatory calculation of ANN combined with M.C., 9311 samples were calculated for a specific steady-state case. The minimum value for the largest bias is 2.4%. For transient case, two tests of different steam-injection conditions were studied. After 1e5 samples, the maximum bias for simulating two transient cases can be limited at 10.7%, and 17.9% respectively.