equinor
Enable prediction runs for FlowNet ensembles of multiple iterations of history matching
Currently FlowNet simulations in prediction mode are only performed for the ensemble of the last completed iteration of the history matching results provided as input. However, there may be cases where it might be desirable to evaluate ensembles from other iterations of the history matching in terms of their performance prediction (e.g. compare prior and posterior ensembles in the prediction period, or compute / visualize RMSE metrics for the accuracy of simulations of various iterations as it is done in the history matching mode). To enable that, the prediction config file should have an (optional) entry for the list of iteration numbers to perform simulations. Then it should be simple to loop over this list and launch ERT to perform prediction simulations of the ensembles of the selected iterations.
If this is pursued, I expect that some of the inputs of the analysis workflow job would need to be adjusted accordingly. For example, the CSV file saving the error metrics of different iterations would have to be passed differently because in the history matching mode the multiple iterations are covered within a single running instance / call of the ERT process, while here we would have multiple ERT instances for a single "iteration" (i.e. ensemble or batch of simulations).