Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Return event state #1300

Merged
merged 14 commits into from
Dec 31, 2020
Merged

Return event state #1300

merged 14 commits into from
Dec 31, 2020

Conversation

rtimms
Copy link
Contributor

@rtimms rtimms commented Dec 15, 2020

Description

UPDATE (16/12/20): the event time and state are now always returned as part of solution.t and solution.y. The solver option for this has been removed.

Add a solver option to return the event time and state as part of solution.t and solution.y.

Also changes the way stepping works in experiments when an event is triggered so that the solution gets back on to the correct reporting period. e.g. if the period is 30s and an event is triggered at 32s we now return the times [0, 30, 32, 60] instead of [0, 30, 32, 62]

Open to suggestions on the option name and on how I've implemented this. Question: should the default be True? I left it as False as typically it seems that solvers don't return the event in this way. Maybe it should be true for experiments at least? I think it is important to get the event time and state back when looking at the solution, especially for doing calculations after the fact.

See here that you now hit the event spot on, regardless of the period

import pybamm

pybamm.set_logging_level("INFO")
experiment = pybamm.Experiment(["Discharge at 1C until 3.6V", "Rest for 1 hour"] * 2)
model = pybamm.lithium_ion.DFN()
sim = pybamm.Simulation(model, experiment=experiment)

solvers = [pybamm.CasadiSolver(), pybamm.CasadiSolver(return_event=True)]
sols = []
times = []
for solver in solvers:
    sol = sim.solve(solver=solver)
    sols.append(sol)
    times.append(sol["Time [s]"].entries)

pybamm.dynamic_plot(
    sols, ["Terminal voltage [V]"], labels=["return_event=False", "return_event=True"]
)

Figure_1

Type of change

Please add a line in the relevant section of CHANGELOG.md to document the change (include PR #) - note reverse order of PR #s. If necessary, also add to the list of breaking changes.

  • New feature (non-breaking change which adds functionality)
  • Optimization (back-end change that speeds up the code)
  • Bug fix (non-breaking change which fixes an issue)

Key checklist:

  • No style issues: $ flake8
  • All tests pass: $ python run-tests.py --unit
  • The documentation builds: $ cd docs and then $ make clean; make html

You can run all three at once, using $ python run-tests.py --quick.

Further checks:

  • Code is commented, particularly in hard-to-understand areas
  • Tests added that prove fix is effective or that feature works

@codecov
Copy link

codecov bot commented Dec 15, 2020

Codecov Report

Merging #1300 (956deb5) into develop (c573c78) will increase coverage by 0.00%.
The diff coverage is 100.00%.

Impacted file tree graph

@@           Coverage Diff            @@
##           develop    #1300   +/-   ##
========================================
  Coverage    98.10%   98.10%           
========================================
  Files          272      272           
  Lines        15229    15236    +7     
========================================
+ Hits         14940    14947    +7     
  Misses         289      289           
Impacted Files Coverage Δ
pybamm/solvers/idaklu_solver.py 89.38% <ø> (ø)
pybamm/simulation.py 96.95% <100.00%> (ø)
pybamm/solvers/base_solver.py 99.03% <100.00%> (+0.01%) ⬆️
pybamm/solvers/casadi_solver.py 99.48% <100.00%> (ø)
pybamm/solvers/processed_variable.py 99.61% <100.00%> (ø)

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update c573c78...956deb5. Read the comment docs.

@rtimms
Copy link
Contributor Author

rtimms commented Dec 15, 2020

one problem this has caused is that doing the extra single step to go from the event time to the next time based on the period adds extra sub solutions. this makes it so that there can be more sub solutions that instructions (e.g. the cccv example now has 19 sub solutions instead of 15, which breaks the custom plot of the discharge).

@rtimms rtimms marked this pull request as draft December 15, 2020 16:39
@valentinsulzer
Copy link
Member

I think return_events should always be True. Actually I don't see the need for the option "False". If events happen, the solution.t you get out will be different from t_eval anyway.
Don't worry too much about sub-solutions, I want to reformat this with a concept of "cycles" anyway

@brosaplanella
Copy link
Member

I agree with Tino. In addition, if you want the solution at the evenly spaced times you provided (which probably will be for storing or further processing), you can always evaluate the processed variables at the desired points.

@rtimms
Copy link
Contributor Author

rtimms commented Dec 16, 2020

thanks, I've removed this as an option and the solvers now always return the event time and state. I've left the step in that gets you back to the times defined by the period based on the initial time. the reason I wanted to do this was incase people were doing e.g. lots of cycles and wanted to use a long period as they are only interested in the solution at the end of the cycle. if you have a long period then when an event is triggered you might get thrown a long way off. at least this way you should be able to trust the solution at the returned times, and aren't getting killed by interpolation error.

@rtimms rtimms marked this pull request as ready for review December 16, 2020 10:52
Copy link
Member

@brosaplanella brosaplanella left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Looks great, thanks Rob!

About the mac tests, I think that this is a GitHub issue (they are doing some maintenance this week which affects mac-os Actions).

@DrSOKane
Copy link
Contributor

I tried merging this into my LithiumPlating branch and got the following error:


AttributeError Traceback (most recent call last)
in
46
47 sim = pybamm.Simulation(model=model, parameter_values=params, experiment=exp)
---> 48 sim.solve()
49 sim.solution.save("/mnt/e/pybamm_data/Plating_1E-8_2E-6_v3.pkl")
50
~/PyBaMM-GEM/pybamm/simulation.py in solve(self, t_eval, solver, check_model, **kwargs)
462 * self._solution.timescale_eval
463 )
--> 464 self.step(dt_event, solver=solver, **kwargs)
465 # Make sure we take at least 2 timesteps
466 npts = max(int(round(dt / exp_inputs["period"])) + 1, 2)
~/PyBaMM-GEM/pybamm/simulation.py in step(self, dt, solver, npts, save, **kwargs)
514
515 self._solution = solver.step(
--> 516 self._solution, self.built_model, dt, npts=npts, save=save, **kwargs
517 )
518
~/PyBaMM-GEM/pybamm/solvers/base_solver.py in step(self, old_solution, model, dt, npts, external_variables, inputs, save)
838
839 # Assign times
--> 840 solution.set_up_time = set_up_time
841 solution.solve_time = timer.time()
842
AttributeError: 'NoneType' object has no attribute 'set_up_time'

@brosaplanella
Copy link
Member

@DrSOKane not sure if this is the cause of the problem you are having, but you would get a similar error if you haven't solved your model yet and ask for the set_up_time. Could you check if your model has been solved?

@DrSOKane
Copy link
Contributor

@brosaplanella I haven't asked for the set up time.

The line raising the error is 840 of base_simulation.py. solution, the object Python is saying is NoneType, is defined on line 837. If solution is NoneType, that implies the integration on line 837 didn't work.

exp_inputs["period"]
- (self._solution.t[-1] - self._solution.t[-2])
* self._solution.timescale_eval
)
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

dt_event can be negative here if switching from a longer period (e.g. 180s) to a shorter period (e.g. w60s), since in that case (self._solution.t[-1] - self._solution.t[-2]) * self._solution.timescale_eval (e.g. 150s) can be longer than the new period even though it is shorter than the old period

@valentinsulzer
Copy link
Member

This error arises when the casadi solver receives a time [t, t+dt] where dt is negative.
There is no check for this currently, so the first thing to add is an error message if dt < 0 in BaseSolver.step.
The place a negative dt is being passed is when correcting to get back to the correct time period here.

To be honest, I'm not sure we should be doing this to get back on the correct time period? e.g. in your [0,30,32,60,90] example, from the point of view of the phase after the event, the times will be [28,30,30,...] which seems a bit weird

@rtimms
Copy link
Contributor Author

rtimms commented Dec 22, 2020

ah good catch, thanks. i’m happy for that step to get to the original period to be removed - it messes with the sub solutions anyway. my thinking was that it would be nice to get the solution at the times you would expect based on starting at t=0 with a given period, and that the event times would just be appended in.

i probably won’t get round to this until the new year, so if someone else wants this to be merged into develop sooner then please go ahead and make the changes

@rtimms
Copy link
Contributor Author

rtimms commented Dec 23, 2020

thanks @tinosulzer ! I’ll revert the cccv script and update the changelog etc. after the holidays

@rtimms rtimms merged commit 66b9589 into develop Dec 31, 2020
@rtimms rtimms deleted the return-event-state branch December 31, 2020 13:34
@valentinsulzer valentinsulzer mentioned this pull request Dec 31, 2020
8 tasks
@valentinsulzer valentinsulzer mentioned this pull request Jan 18, 2021
4 tasks
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

4 participants