A reliability analysis method based on quantile regression and feedforward neural network

The surrogate model-based reliability analysis method has drawn researchers' studies in recent years. Deep learning is a powerful method to construct a surrogate model. Due to the data's existing noise which will lead to aleatoric uncertainty, this paper proposes the Monte Carlo (MC) quantile regression method to build the surrogate model based on quantile regression and feedforward neural network. During the training process, the quantile levels for different inputs are randomly changed in each iteration. In the prediction stage, the same input is attached with varying quantiles to quantify the aleatoric uncertainty, and the input is attached with quantile level 0.5 to predict the output. Two examples are used to validate the effectiveness of the proposed MC quantile regression method. The results show that the MC quantile regression method can accurately quantify the aleatoric uncertainty and construct an accurate surrogate model to analyze system reliability.