Allow for multiple separated network in a single FlowNet

wouterjdb commented 3 years ago

This issue is to support a stepwise implementation of a layered FlowNet approach. This issue describes a feature that allows for the generation of 2 or more independent network in a single FlowNet.

What Allow for multiple networks to be generated in a single FlowNet that are only communicating via well connections. This can be achieved relatively straight forward with the current code and can serve as a proof of concept that history matching performance could be improved by a layered approach.

How

The first three tasks are addressed in issue #277, PR #279.

[x] Allow to specify layering in the FlowNet configuration file, for example:

flownet:
data_source:
    simulation:
        layering:
           - [ 1, 3 ]
           - [ 4, ... ]

[x] Translate the layering for each well into depth intervals. The depth intervals are what will need to be used to assign layer_id's to each well connection (this to propagate simulation specific information in the code).
[x] Create a function that assign layer_id's to well connections. For example, this could be added to field_data.well_connections as an additional column.

This task is addressed in PR #287.

[x] A function needs to be created that allows for creating the _grid_cell_bounding_box for a specific layer. This can be done by extending this function: https://github.com/equinor/flownet/blob/d00960b44c561e2ce5e5630f278179b25ac59241/src/flownet/data/from_flow.py#L283

[x] Update perforation strategy PR #296

PR #304.

[ ] In __generateconnections.py the create_connections() could be expanded to allow for splitting models based on depth intervals per well. That is, an additional input "layer_id" could be supplied that identifies which well coordinates in df_coordinates belong to which zones.
[ ] By using this information, as well as updating the concave_hull_bounding_box, one could now run the _generate_connections function multiple times (which in turn performs Mitchells Best Candidate and Delaunay triangulation). This will generate multiple lists of start and end points for all connections in the FlowNet. Some rules for dividing the additional nodes over the various network need to be used (for example divide based on number of connections associated with a layer_id).

tayloris commented 3 years ago

In issue #171, we discuss the possibility to influence the network generation using flow diagnostic. Maybe we could an option to indicate which wells will communicate with each layer. In that way, the nodes that are created in that layer will connect with the wells specified by the user.

wouterjdb commented 3 years ago

In essence we do that based on the geological formations where wells are perforated. However, this of course does not give us communication wells due to juxtaposition.

For now, however, we are taking the simple step of fully disconnected networks. A next step should be to include in a better way faults, and juxtaposition, and there information from flow diagnostics could be incorporated.

equinor / flownet

Allow for multiple separated network in a single FlowNet #276

[x] Create a function that assign layer_id's to well connections. For example, this could be added to `field_data.well_connections` as an additional column.

equinor / flownet

Allow for multiple separated network in a single FlowNet #276

[x] Create a function that assign layer_id's to well connections. For example, this could be added to field_data.well_connections as an additional column.

[x] Create a function that assign layer_id's to well connections. For example, this could be added to `field_data.well_connections` as an additional column.