I want to creat RBDO model (based on metamodel) without considering any marginal for environmental parameters. In fact, I have only a data set for environmental parameters. I read “RBDO_manual”, but I couldn’t find anythings about my problem, How to build RBDO model without defining marginals for environmental parameters?
This use case is currently not implemented in UQLab. The module requires the distribution of the environmental variables to:
either sample them (in simulation-based methods) or to do the appropriate transform to the standard Gaussian space (for FORM-based or decoupled methods);
construct the space where the surrogate model is built.
How many data points do you have?
A workaround I could suggest to you is:
Learn the marginal distributions of the environmental variables using the Input Inference module (you can find an example here) and then use those for the RBDO analysis
Use your data to externally build a surrogate model. Since you do not have a model you cannot do active learning (AK-MCS type method), so you can only rely on a unique surrogate model built prior to your analysis. Once you have such a surrogate model (and you are sure it is accurate enough for the purpose of your analysis) you can pass it to the RBDO analysis as default model and make sure the construction of the surrogate within the analysis is disabled.
I hope I am answering your questions. If not please let me know.