Application of Quantum Circuit Learning for Statistical Safety Evaluation

Abstract

To reduce the computational demand in the best estimate plus uncertainty (BEPU) analysis, an accurate and inexpensive machine learning model is expected to be used to replace the high-fidelity RELAP5 code for rapid determination of the uncertainties on the figure of merit of interest. One of the problems associated with the application of machine learning is overlearning. Quantum circuit learning is the quantum analogue of classical deep learning, which is expected to be less prone to overlearning because the optimized parameters are bound by unitary transformations in the quantum circuit. In this paper, quantum circuit learning is applied to the BEPU analysis of the fuel peak cladding temperature (PCT) for a small-break LOCA scenario in PWRs. The parameterized quantum circuit is trained using a small number of the RELAP5 analysis results and the prediction accuracy of the 95th percentile value of the PCTs is investigated. By taking a large number of layers of parameterized quantum circuits depending on the complexity of the analysis target and applying a global optimization method to optimize the parameters, the 95th percentile value of the PCTs predicted by the quantum circuit learning is found to result in better accuracy and smaller variability than linear quadratic regressions.