Wombat

- Removes the pre-public release analysis data and results, located at`library/original-outdated`, from the repository, but can be found in all releases prior to v0.9. This is being done to manage the package size and ensure it can be published to PyPI.

Bug Fixes

- Repairs that had a weather delay extended repairs past the end of the shift now properly extend the repair into future shifts instead of completing early. This has a small negative impact on availability because that means some repairs can take longer than they had originally.
- Traveling to a system for a repair where the timing extends beyond the end of the shift, but into the next shift, is now registered as a shift delay just like travel weather delays that extend beyond the end of the current shift but before the start of the next shift. This has a small positive impact on availability because a turbine or cable may not start being repaired until weather is more consistently clear, rather than starting it and extending it for many shifts.
- `Windfarm.cable()` now correctly identifies 2 and 3 length cable naming conventions to differentiate which version of the cable id is being retrieved.
- An edge case where events occurred just after the end of a simulation has been resolved by checking the datetime stamp of that event and not adding any action log to the simulation that is after the `WombatEnvironment.end_datetime`.
- A bug in how the total wind farm operating level was calculated is updated to account for substation downtime, rather than using a sum of all turbine values.
- `Metrics.time_based_availability` and `Metrics.production_based_availability` have been updated to use to take advantage of the above fix. Similarly, the time-based availability skews higher now, as is expected when taking into account all availability greater than 0, and the energy-based availability drops moderately as a result of accounting for the substation downtime.

General Updates

- `Metrics.equipment_labor_cost_breakdowns` now has a `by_equipment` boolean flag, so that the labor and equipment costs can be broken down by category and equipment. Additionally, `total_hours` has been added to the results, resulting in fewer computed metrics across the same set of breakdowns.
- "request canceled" and "complete" are now updated in the logging to directly state if it's a "repair" or "maintenance" task that was completed or canceled to ensure consistency across the logging messages. As a result, `Metrics.task_completion_rate()` can now correctly differentiate between the completed tasks effectively.
- The use of unique naming for the servicing equipment is now enforced to ensure that there is no overlap and potential confusion in the model.
- New, experimental plotting functionality has been added via `wombat.utilities.plot`.
- `plot_farm_layout` plots the graph layout of the farm. Note that this will not work if realistic lat/lon pairs have not been provided in the layout CSV.
- `plot_farm_availability` plots a line chart of the monthly overall windfarm availability. Additional toggles allow for the plotting of individual turbines in the background and/or a 95% confidence interval band around the farm's availability
- `plot_detailed_availability` plots a heatmap of the hourly turbine and farm operational levels to help in debugging simulations where availability may be running suspiciously low (i.e., a turbine might have shut down or a cable failure didn't get repaired within a reasonable time frame).
- `Simulation.service_equipment` is now a dictionary to provide clear access to the servicing equipment details.
- `Simulation.from_config()` requires both a library input and a file name input for the configuration after deprecating the `library` field from the configuration requirements due to it being cumbersome
- `Metrics.opex()` provides the opex output by category if `by_category` is set to be `True`

Methodology Updates

- Subassemblies and cables are now able to resample their next times to failure for all maintenance and failure activities, so that replacement events reset the timing for failures across the board.
- Systems are no longer interrupted once the servicing equipment addressing a repair or maintenance task is dispatched, and instead are interrupted just before the crew is transferred to begin repair, maintenance, or unmooring. This has a small net positive increase in the availability for short travel distances, but can vary if the travel time is more than a few hours (i.e., large distance to port, slow travel due to weather, & etc.)
- When a tow to port repair is in the queue, no other repairs or maintenance activities will occur for that turbine until it's brought into port for repairs.

Deprecations

- The original library structure has been fully deprecated, please see the Reference/How To example for more details on converting data that has not yet been updated.
- All PySAM functionality has been deprecated as it's no longer relevant to this work.
- The `Simulation.config` no longer requires the `library` field to reduce conflicts with sharing data between users.

- Fixes a bug where servicing equipment waiting for the next operational period at the end of a simulation get stuck in an infinite loop because the timeout is set for just prior to the end of the simulation, and not just after the end of the simulation's maximum run time.

Bug Fixes

- Most of the ` type: ignore` comments have been removed, or the past errors have been resolved
- Failure and maintenance logic in the `Cable` and `Subassembly` models have been wrapped in `if`/`else` blocks to ensure previously unreachable code still can't be reached under limited conditions
- Fixes a bug in where the mobilization duration is logged. Previously, the mobilization duration was logged upon completion, but now is logged while during the actual mobilizing, with costs still being logged upon completion.
- Fixes an uncommon error where repairs are made twice causing the simulation to fail. This was caused by a control mismatch in the unscheduled repair process, and was fixed by better tracking and controlling when a repair moves from "pending" (submitted), to "processing" (handed off to the servicing equipment for repair or to the port for a tow-to-port repair cycle), to "completed" (the repair is registered as complete). Additionally, the servicing equipment no longer relies on receiving the first repair from the repair manager on dispatch, and follows the same `get_next_request()` logic that will occur following the initial dispatch and repair.
- Fixes an error where a repair has `replacement=True` and `operation_reduction` < 1 both resets the `operating_level` back to 1 and readjusts for the `operation_reduction`, which can cause `operating_level` > 100%
- `ServiceEquipment.crew_transfer()` uses a while loop in place of an `if` with recursion strategy to handle unsafe transfer conditions to continuously wait until the next shift's available window.

Features

- Adds a `non_stop_shift` attribute to `ServiceEquipmentData`, `UnscheduledServiceEquipmentData`, `ScheduledServiceEquipmentData`, and `PortConfig` that is set in the post-initialization hook or through `DateLimitsMixin._set_environment_shift()` to ensure it is updated appropriately. Additionally, all checks for a 24 hour shift now check for the `non_stop_shift` attribute.
- `Metrics.emissions()` has been added to the list of available metrics to calculate the emissions from idling at port or sea, tranisiting, and maneuvering. Co-authored by and inspired by analysis work from hemezz.
- `Simulation` now accepts a `random_seed` or `random_generator` variable to seed the random number generators for Weibull failure timeouts and wait timing between event completions. Setting the `random_seed` to the same value from one simulation to the next will net the same results between different simulations, whereas the `random_generator` can be used to use the same generator for a batch of simulations.

General

- All `assert` statements are now only called when type checking is performed
- Replaces all `.get(lamda x: x == request)` with a 10x faster `.get(lambda x: x is request)` to more efficiently filter out the desired event to be removed from the repair manager and port repair management.
- `WombatEnvironment.weather` is now a Polars DataFrame to improve efficiency and indexing bottlenecks introduced in Pandas 2.0.
- All subassembly cable files are read in once, and stored in a dictionary to provide a modest speed up for the simulation initialization.

- Features
- `Metrics.process_times()` now includes the time_to_start, representing the time between when a request is submitted, and when the repairs officially start.
- Expand the acceptable date formats for the weather profiles to allow for year-first.

- Replace Flake8 and Pylint in the pre-commit workflow with ruff, and fix/ignore the resulting errors as appropriate
- Features:
- Weather data now has the ability to contain more than just the required "windspeed" and "waveheight" columns. This will allow for easier expansion of the weather model in the future, and increase compatibility with other NREL techno economic modeling frameworks.
- Bug fixes:
- Maintenance and failure simulation process interruptions were occuring prior to starting the process timing, and causing simulation failures.
- Duplicated parameters were being processed in `WombatEnvironment.log_action` stemming from improper handling of varying parameters in some of the more complex control flow logic in *in situ* repairs.
- Another edge case of negative delays during crew transfers where there is insufficient time remaining in the shift after account for weather, so the method was called recursively, but not exiting the original loop.
- `Port` management of *in situ* and tow-to-port capable tugboats wasn't properly accounting for tugboats of varying capabilities, and assuming all tugboats could do both. The vessel management and repair processing were out of sync causing duplicated turbine servicing/towing.
- `RepairManager` has consolidated the dispatching of servicing equipment to be the following categories instead of the previously complex logic:
- If a request is a tow-to-port category, have the port initiate the process
- If the dispatching thresholds are met, then have the port initiate a repair for port-based servicing equipment, otherwise the repair manager will dispatch the appropriate servicing equipment.
- `ServiceEquipment.weather_delay()` no longer silently processes a second weather delay.
- Intra-site travel is now correctly logging the distance and duration of traveling between systems at the site by moving the location setting logic out of `crew_transfer` method and being entirely maintained within, or next to, the `travel` method of `ServiceEquipment`.
- `RepairManager` now properly waits to dispatch servicing equipment until they are no longer under service from another servicing equipment unit.
- Servicing equipment are no longer simultaneously dispatched from both the `Port` and `RepairManager` causing simulations to potentially error out when the `System.servicing` or `Cable.servicing` statuses are overridden by the second-in-line servicing equipment. This was resolved by the port waiting for the turbine, a tugboat, and a spot at port to become available, then immediately halting the turbine, and starting the tow-in logic.
- Missing `ServiceEquipment.dispatched` status updates have been amended, so no matter the operation, a piece of servicing equipment should be set to `dispatched = True` until the crew is back and repair is completed.
- To avoid multiple dispatches to a single turbine, cable, or substation, or multiple dispatches of a single vessel/onsite equipment, a random timeout between 0 and 30 seconds (in simulation time) is processed, then a status double-checking occurs.
- `Metrics.process_times()` now includes the "N" column to indicate the number of processes that fall into each category.