An improved uncertainty calibration approach for reliable satellite temperature interval prediction

Accurate satellite temperature interval prediction is critical for satellite reliability analysis and health state evaluation. However, existing uncertainty quantification methods often yield imprecise uncertainty estimates for the complex task of predicting satellite temperatures. Therefore, enhancing the quality of uncertainty quantification by recalibrating uncertainty results is of paramount importance. This paper introduces an advanced model uncertainty calibration approach for satellite temperature interval prediction, leveraging the Monte Carlo Dropout and quantile calibration methods. Specifically, the Monte Carlo Dropout method is employed to quantify model uncertainty by generating output distribution samples. Concurrently, the uncertainty quality is enhanced by adjusting the quantile levels using the quantile calibration method, resulting in a more accurately calibrated prediction interval. The efficacy of the proposed method is validated through a simulation example and a real-world engineering case, demonstrating that the calibrated prediction interval coverage probability aligns more closely with the specified confidence levels, thereby rendering the predicted uncertainty interval more credible.