multiple well gradients

[zhen0z]

Hi,

I'm working on using the sensitivity functionality and wanted to check if it's possible to compute gradients using multiple wells. The original example is as follows:

gradient

function objective_function(model, state, Δt, step_i, forces)
grat = JutulDarcy.compute_well_qoi(model, state, forces, :Producer, SurfaceGasRateTarget)
return Δtgrat/(inj_ratesum(dt))
end
data_domain_with_gradients = JutulDarcy.reservoir_sensitivities(case, result, objective_function, include_parameters = true)

is it possible to do somthing like this:

gradient

function objective_function(model, state, Δt, step_i, forces)
grat = JutulDarcy.compute_well_qoi(model, state, forces, [:Producer1,:Producer2], [SurfaceGasRateTarget,SurfaceGasRateTarget])
return Δtgrat/(inj_ratesum(dt))
end
data_domain_with_gradients = JutulDarcy.reservoir_sensitivities(case, result, objective_function, include_parameters = true)

Thanks,
Zhen

[moyner]

Yes, this should be possible. Try something like this:

function objective_function(model, state, Δt, step_i, forces)
    grat = 0
    for w in  [:Producer1,:Producer2]
         grat += JutulDarcy.compute_well_qoi(model, state, forces, w, SurfaceGasRateTarget)
    end
    return Δt*grat/(inj_rate*sum(dt))
end
data_domain_with_gradients = JutulDarcy.reservoir_sensitivities(case, result, objective_function, include_parameters = true)

I have not tested this code, but if you can find the ECMOR paper there is an example for the Norne model where we do this kind of setup over all producers.

[zhen0z]

Thanks, that's very helpful, and I have another two small problems:

First, if I have some observation data, should I do this:
function objective_function(model, state, Δt, step_i, forces)
grat = 0
for w in [:Producer1,:Producer2]
grat += JutulDarcy.compute_well_qoi(model, state, forces, w, SurfaceGasRateTarget)
end
grat = grat .- observation_data
return Δtgrat/(inj_ratesum(dt))
end

Second problem, I'd like to set the production rate to zero and use the bottom hole pressure (BHP) of the producer as the objective function, treating it like a monitoring well. Could I do that?

Thanks!

[moyner]

Your observation data looks good to be - but you may have to use step_i to pick the right observations for that time-step (unless your observed rates are constant). The objective function is called for each report step in the simulation.

You can set DisabledControl instead for the well to treat it as an observation well. The BHP will still be updated.

Hope this helps!

[zhen0z]

Thanks for the explanation, use DisabledControl is straightforward.

However, for the observation data, let's say I have a time series surface gas production rate observation, and I want to minimize the summation of the differences between the simulated surface gas production rate and the observed surface gas production rate. how should I incorporate it?

[moyner]

It is a bit tricky to explain without a full script, but if your case has time steps passed as dt and dt has length 10 then step_i passed to your objective will vary from 1 to 10. You can use this index to pick the correct observations from however you have it stored, for example by making a function get_gas_rates_for_step(step_i) and calling it inside the objective function.

[zhen0z]

Thanks, it seems that I can just simply pass the time steps into the objective function, and it works well. Thanks for your help, it's a very nice package! Best regards, Zhen Olav Møyner ***@***.***> 于2024年9月24日周二 18:25写道：

…

It is a bit tricky to explain without a full script, but if your case has time steps passed as dt and dt has length 10 then step_i passed to your objective will vary from 1 to 10. You can use this index to pick the correct observations from however you have it stored, for example by making a function get_gas_rates_for_step(step_i) and calling it inside the objective function. — Reply to this email directly, view it on GitHub <#60 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/BIOR2UWASKMY5GGLZ6UKQYDZYGOCNAVCNFSM6AAAAABOWOAHFKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDGNZRHA4DSNBXHE> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[moyner]

That's great! I'll close this issue now. Feel free to reopen or add another issue if you have follow up questions.