For large cases (many wells, long history) the construction of a FlowNet case from scratch might take quite some time. For repeated experiments in which only history match settings are varied, significant speed-up could be obtained by re-using an existing architecture and schedule. These could be read from stored pickled files generated during a previous experiment. This functionality is already used when running FlowNet in prediction mode and could be made available also for experiments in history match mode.
For large cases (many wells, long history) the construction of a FlowNet case from scratch might take quite some time. For repeated experiments in which only history match settings are varied, significant speed-up could be obtained by re-using an existing architecture and schedule. These could be read from stored pickled files generated during a previous experiment. This functionality is already used when running FlowNet in prediction mode and could be made available also for experiments in history match mode.