Interval predictor model
Interval Predictor Models (IPMs) are a class of meta-model which describe the expected spread of the output of a model whilst making very few assumptions about the model. The meta-model finds an upper and lower bound for the predicted output of the model using a function chosen by the user. Therefore, once trained, the meta-model can make predictions very quickly. Interval Predictor Models offer a robust quantification of uncertainty, even when few data points are available. This makes them useful when a very computationally expensive model limits the number of simulations which can be performed.
Creating an IPM of the performance function yields bounds on ( and ). Hence we obtain bounds on the failure probability by Monte Carlo simulation for and .
Contents
What do I need to train an Interval Predictor Model?
The minimal implementation requires at least one RandomVariable defined in the Input Object, and one Model.
How do I perform a reliability Analysis with an Interval Predictor model?
In the standard Reliability Analysis workflow the Model should be replaced with a Model containing an Evaluator, in turn containing an @IntervalPredictorModel.
Where are my results?
The results of the analysis are stored in a FailureProbability object
How is the Interval Predictor Model object constructed?
See the page of @IntervalPredictorModel object for more details
Theory A brief explanation of the design of Experiment methods is available on our wikipedia article |