Fix storage optimisation algorithm

[noracato]

New implementation of output efficiency, so it is not applied twice after calculating the curve.

Finally closes #1300

[thomas-qah]

Hey @noracato! I checked out the battery-opt branch and ran rspec for it locally. Doing this returned 5 errors:

3205 examples, 5 failures, 20 pending

Failed examples:

rspec ./spec/models/qernel/merit_facade/storage_optimization_spec.rb:216 # Qernel::MeritFacade::StorageOptimization#reserve_for with a single battery with an output efficiency of 0.8 calculates the battery reserve
rspec ./spec/models/qernel/merit_facade/storage_optimization_spec.rb:223 # Qernel::MeritFacade::StorageOptimization#reserve_for with a single battery with an output efficiency of 0.8 calculates the battery load
rspec ./spec/models/qernel/merit_facade/storage_optimization_spec.rb:238 # Qernel::MeritFacade::StorageOptimization#reserve_for with a single battery with an output efficiency of 1.2 calculates the battery reserve
rspec ./spec/models/qernel/merit_facade/storage_optimization_spec.rb:245 # Qernel::MeritFacade::StorageOptimization#reserve_for with a single battery with an output efficiency of 1.2 calculates the battery load
rspec ./spec/requests/api/v3/saved_scenarios_spec.rb:107 # Saved scenarios API when fetching a single saved scenario when the saved scenario exists but the scenario does not has no data for the scenario

These seem to correspond with the error messages that semaphore currently returns (https://quintel.semaphoreci.com/jobs/059c9c4f-9bff-4e8b-a130-c630c114057d/plain_logs.txt)

Let's discuss this thursday/friday :)

[mabijkerk]

The difference between running the spec locally on Mac with respect locally on Linux or on Semaphore was caused by an operating system difference in the sort method of Ruby. Non-unique values are sorted in a different way by both operating systems. To counter this we have implemented a second-level attribute to the sort function, that sorts on the index as well, and therefore guarantees a stable result.

[mabijkerk]

My test case in #1300 now produces expected results.

Blank nl2019 scenario with the following changes:

capacity_of_energy_flexibility_flow_batteries_electricity: 500.0
capacity_of_energy_power_solar_pv_solar_radiation: 15000.0
capacity_of_energy_power_wind_turbine_inland: 15000.0
settings_enable_storage_optimisation_energy_flexibility_flow_batteries_electricity: 1.0

With the following queries:

EACH(
  V(energy_flexibility_flow_batteries_electricity,"number_of_units * typical_input_capacity"),
  MAX(V(energy_flexibility_flow_batteries_electricity,electricity_input_curve))
 )

EACH(
  V(energy_flexibility_flow_batteries_electricity,"number_of_units * electricity_output_capacity"),
  MAX(V(energy_flexibility_flow_batteries_electricity,electricity_output_curve))
 )

Obtains the following results:

[
  500.00000000000006,
  500.00000000001455,
]

[
  250.00000000000003,
  250.0000000000036,
]

This is as expected: it means that the output capacity no longer has an additional cap. The output capacity realized by the algorithm is the installed output capacity and the input capacity realized is the installed input capacity. This means that we can merge this PR and the corresponding PR on quintel/merit#153.

@noracato as the results above show I'm still seeing the floating point error addressed in #1331.

[mabijkerk]

🔋