errors associated with capacity predictions are determined to be sufficiently low. This study suggests using a recursive strategy that builds upon the strengths of Bayesian statistical inference to assess the upper bounds of forecasting errors. In our approach, measurements are used to (1) improve the models through calibration, (2) assess the usefulness of the simulation model and (3) determine the forecasting errors. These three merits of the approach described in this paper are noteworthy. Improve the models through calibratio The strength of the adapted Bayesian calibration tool, GPM/SA comes from the inherent consideration of the parameter uncertainties (known unknowns) and the model form errors (unknown unknowns). For the case studies discussed herein, the prior probability distributions of the parameters are treated as uniform probability distributions. However, GPM/SA is not limited to uniform prior distributions and has the a priori capability to operate on other forms of probability density functions. Assess the usefulness of the simulation model If the physics or engineering phenomena modeled in a simulation model is not refined enough for certification purposes, the model fidelity cannot be improved through companion experimental campaigns and calibration efforts. Therefore, we consider assessing the usefulness of a simulation model for forecasting as the first necessary step in certification. For a given a simulation model and a companion set of experimental measurements, we formulated a simple approach to determine the usefulness of a simulation model to deliver solutions with sufficient fidelity. Determine the forecasting errors By exploiting the available experiments, we estimate an upper bound for forecasting errors. The forecasting error is considered as a metric providing a notion for the effects of uncertainty within the domain of applicability. Benefits of forecasting metric are two fold, for a given experimental budget the forecasting metric can quantify the errors associated with forecasting, secondly the forecasting metric will provide when the sufficient number of experiments have been performed to achieve desired level of forecasting fidelity (quantified error level). 444
RkJQdWJsaXNoZXIy MTMzNzEzMQ==