Skip to content

Latest commit

 

History

History
790 lines (744 loc) · 49.1 KB

CHANGELOG.md

File metadata and controls

790 lines (744 loc) · 49.1 KB

Switch 2.0.9

This release includes several new features, most notably implementation of pumped storage hydro. It also renames one output file and changes the balancing rules for hydro systems when there are multiple timeseries in the same investment period.

The updates are summarized below. For more details, see the git commit log.

Changes that may affect existing models and results

  • The switch_model.generators.extensions.hydro_system module now tracks hydro reservoir levels separately for each timeseries instead of linking them together across the whole period.
    • Previously the hydro_system module tracked levels of reservoirs across all the timepoints in a period, as if each timeseries was linked to the one after it. This was an error, and in models with multiple timeseries with different weights, Switch could retain water during low-weight timeseries, then release it during high-weight timeseries, effectively producing free energy.
    • The new approach matches the general principle in Switch that each timeseries is independent and it must be possible to repeat each timeseries many times in a row if necessary. Although this is more correct than the previous approach, it is also more conservative: hydro networks must now reach the same level (or a prespecified level) at the start and end of each timeseries, instead of the start and end of each period.
    • Users should now set res_initial_vol and res_final_vol for each timeseries in reservoir_ts_data.csv instead of setting initial_res_vol and final_res_vol for each period in reservoirs.csv. The standard upgrade script will automatically rename and move these columns if needed. It will also interpolate between the previous full-period reservoir volumes to set the reservoir volumes at the start and end of each timeseries.
    • For models with a single, long timeseries for each period (the most common use case for this module), this change won't affect results. However, for models with multiple timeseries in each period with different weights, this update may change the model results.
    • The res_initial_vol and res_final_vol inputs are now optional. If res_initial_vol is set but not res_final_vol, Switch will set the final level equal to the initial level. If neither is set, Switch will choose an optimal initial volume for each timeseries and also return to that level at the end of the timeseries. (For most models, it is best not to set res_final_vol, so Switch will return the reservoir to the starting level. This makes it possible to have an arbitrary number of repetitions of each timeseries.)
  • The gen_project_annual_summary.csv output file has been renamed to dispatch_gen_annual_summary.csv. The new name is more consistent with the other dispatch summaries, such as dispatch_zonal_annual_summary.csv or dispatch_annual_summary.csv.

New features

  • The switch_model.generators.extensions.hydro_system module can now model pumped storage hydro systems.
    • To use this feature you must also add switch_model.generators.extensions.storage somewhere above switch_model.generators.extensions.hydro_system in modules.txt.
    • Hydro generators can be designated as reversible (able to do pumped hydro storage) by setting gen_storage_efficiency in gen_info.csv to a numeric value (generally 0.5 - 1.0).
    • When generators are identified as reversible, water can be pumped from the node below the generator to the one above, with the specified round-trip efficiency. The nodes above and below the generator should have reservoir data in reservoirs.csv to indicate the amount of storage available. If you currently have a sink node below the dam, you may need to add a reservoir with a fixed size for the lower pool, then connect the sink node downstream of that.
    • The storage module now defines a new set called ALL_STORAGE_GENS, which will rarely be used. This is the union of pumped-storage hydro generators and standard storage generators (usually batteries), which are still listed in STORAGE_GENS.
  • If you have installed pulp (easy to do via pip or conda), you can use the cbc solver bundled with it by specifying --solver pulp_cbc when running Switch. CBC is a fairly fast solver and can be difficult to install on Windows, so this may be a convenient way to use it.
  • The switch_model.balancing.diagnose_infeasibility module now relaxes bounds on variables in addition to constraints. These will show up in the reporting as VariableName_bounds, and can be turned back on via --no-relax VariableName_bounds. This can give a clearer picture of which constraints and/or bounds are interacting to make a model infeasible.
  • The switch_model.hawaii.ev module now balances EV charging by date instead of timeseries, if the tp_date column is provided in timepoints.csv. This forces full charging every day when using multi-day timeseries, rather than allowing EVs to delay charging multiple days, possibly exceeding their storage capacity.

Switch 2.0.8

This release includes several new features, bug fixes and compatibility improvements. It should have minimal effect on results from existing models other than some improvements and bug fixes in the electricity_cost.csv output file.

The updates are summarized below. For more details, see the git commit log.

Changes that may affect existing models and results

  • In Switch 2.0.7 and earlier, the SystemCostPerPeriod_Real column in outputs/electricity_cost.csv was a (mislabeled) annual cost but SystemDemand_MWh was a total for the whole period. The EnergyCostReal_per_MWh was calculated as the ratio of these, making it too low by a factor equal to the number of years in the period. Starting with Switch 2.0.8, we report both costs and quantities on a per-year basis (SystemCostPerYear_Real and SystemDemandPerYear_MWh), and the error in EnergyCostReal_per_MWh has been corrected.
  • Pyomo dependencies have been updated to to versions 6.0.0-6.7.1 (the current latest version).
    • If you are using Pyomo 5.7, you will need to update to at least 6.0. This change was made to avoid complex dependencies on pyutilib.
  • Made --no-save-solution the default behavior, so the large results.pickle file will not be written unless requested. Users who have previously set this flag should now remove it. Users who need the results.pickle file (sometimes useful for reloading and inspecting a previous solution, with --reload-prior-solution) should add the new --save-solution-file flag.
  • Fixed a bug in hawaii.save_results that erroneously grouped all generators into the same zone in capacity_used_by_technology.csv and production_by_technology.csv
  • build_gen.csv will now contain a . instead of 0.0 for periods when generators can't be built
  • Fixed a bug that previously crashed Switch when using the gurobi or cplex solver without any suffixes.
  • Fixed a bug that previously dropped users into a debugging terminal when an infeasible model was encountered.
  • Fixed a bug that crashed hawaii.save_results if the model did not use the transmission.local_td module
  • Fixed bugs in balancing.diagnose_infeasibility that would crash when analyzing singleton constraints or constraints formulated as (lower_bound, expression, upper_bound).

New features

  • Added a hydrogen production module. This produces a fuel (by default called "Hydrogen") that can be used by any generator.
    • There is assumed to be one set of hydrogen production equipment per load zone, and one possible cost for each component, regardless of location. Hydrogen production equipment, once built, continues to the end of the study (similar to transmission infrastructure).
    • This feature can be turned on by adding switch_model.energy_sources.hydrogen.production to your module list and adding parameters for each production component to inputs/hydrogen.csv.
    • Components include
      • electrolyzer (cost per MW power rating and per kg produced, kg of hydrogen produced per MWh of power consumed, lifespan),
      • liquefier/refrigerator (capital and fixed costs per kg/hr of capacity, variable cost per kg produced, MWh of electricity required per kg liquefied, lifespan) and
      • liquid storage tank (cost per kg of capacity, lifespan).
    • The module is designed to represent storage of liquefied hydrogen in NASA-type above ground tanks, but it can be used to represent compressed hydrogen storage instead by setting the liquefier energy requirement and cost to zero, so compressed hydrogen can be stored as if there were no cost for liquefication.
    • This module assumes production within the day can be used without a tank, but any surplus or shortfall for the day must be moved to a tank, and the tank must be large enough to hold all the withdrawals for the whole year. This formulation will work regardless of the actual order of removals, so it avoids the need to link sequences of days. However, if tank costs are high, it may be overly conservative.
    • If using this module with multi-day timeseries, you should fill in date markers in the td_date column of inputs/timepoints.csv (see below).
    • An example implementation can be found in examples/hydrogen
    • Future work could extend this to allow different costs for different locaitons and vintages, retire components at end of life, allow multiple production facilities per zone, better represent compressed (not liquefied) storage and consider the sequence of dates in the year when setting the tank size.
  • Added option to perform optimal early retirement and suspension for generators
    • Setting gen_can_retire_early or gen_can_suspend flags in gen_info.csv to True or 1 now allows generators to be retired permanently or temporarily suspended before they reach their maximum age. In both cases, the amount of capacity of a particular vintage that is offline in a given period will be shown by the new SuspendGen variable. These both default to False/0, and gen_can_suspend will take precedence (temporary suspension will be allowed) if both are set to True/1.
    • The GenCapacity expression now reports only capacity that has not been suspended, and a new column in gen_cap.csv, SuspendGen_total, shows the total capacity in each generation project that is suspended/inactive in each period.
    • Suspended generators avoid fixed O&M costs, but must continue to pay capital recovery (amortized capital costs) as normal.
  • Added generator retrofit capability via a new switch_model.generators.extensions.retrofit module. This makes it possible to define retrofit generators that supersede existing generators.
    • Retrofit options are implemented by defining a new generation project that can replace a previously built one (i.e., it performs like the original generator plus the retrofit) and adding columns to gen_retrofits.csv showing all allowed combinations of base projects that can be replaced by retrofit projects. In each row, base_gen_project shows the name of the original (base) project and retrofit_gen_project shows the name of a retrofit project that can replace it.
    • Retrofit projects will only be built if the base project has also been built in the same or an earlier period. When a base project is retrofitted, the base project is suspended (via SuspendGen) and the retrofit version is built and operated instead. In addition, retrofit projects are automatically suspended at the end of life of the base project. (To enable these behaviors, gen_can_retire_early or gen_can_suspend must be set to True or 1 in gen_info.csv for both the base project and the retrofit version.)
    • Because of this framing, retrofitted projects will have capital expenditure equal to the capital recovery for the base project plus capital recovery for the retrofit project. So gen_overnight_cost for the retrofit project should be set equal to the cost of the retrofit work, not the combined project. However, fixed and variable O&M will no longer be collected for the base project, so O&M cost inputs for the retrofit project should be the ones that apply for the total retrofitted project.
    • Capital recovery for the retrofitted project will be amortized over the remaining life of the base project that it replaces, which may cause faster capital recovery than would otherwise be expected for these assets.
    • A working example is available in examples/retrofits
  • Added tp_date parameter to identify which date each timepoint falls on. This is used by modules that enforce constraints between hours of the same date (e.g., hydrogen production or intra-day demand response). It defaults to be the same as the timeseries (the previous behavior), which works fine if each timeseries is one day long. But if models use multi-day timeseries, they should specify a code for the corresponding date for each timepoint in the tp_date column of inputs/timepoints.csv.
  • Added --retire {early, mid, late} command-line flag. This flag controls whether to retire projects at the start of the period when they reach end-of-life ('early') (i.e., only run if they survive to the end of the period), or retire them if they survive past the middle of the period ('mid'), or extend operation to the end of the period when they reach end-of-life ('late'). Late is the default and matches previous behavior by Switch. Early and mid match some other models' behavior. Note that early requires that a project's life extends not just up to the end of the period, but beyond it into the next period, or else it will retire at the start of the period.
  • Added a new, optional gen_storage_energy_fixed_om parameter to gen_build_costs.csv to indicate fixed O&M costs for the energy component of storage projects. Note that no column is needed for variable O&M for the energy component, because this is already covered by gen_variable_om for the power component.
  • Switch now allows negative values for fixed and variable O&M for generators. These can be useful for representing subsidies that produce net-negative carrying or operating costs.
  • Switch is now compatible with appsi_highs and other appsi_* solvers.
    • HiGHS is one of the fastest open-source solvers we have found for medium-sized models.
      • To use HiGHS, first install Switch following the standard instructions, then install HiGHS itself into your environment via either a binary download or conda install highs, then install the HiGHS-Python bindings into your environment via pip install highspy, then run with --solver appsi_highs. (It is also possible to download an AMPL-compatible version of HiGHS from https://portal.ampl.com/user/ampl/download/highs, place it in your path and then set --solver highs.)
      • appsi_highs requires Pyomo 6.4.3 or later. The ampl version of HiGHS can be used with any version of Pyomo.
    • CBC is another of the fastest open-source solvers we have found for medium-sized models. It can be installed on Linux and macOS via conda install coincbc, then use --solver cbc or --solver appsi_cbc. It currently has limited availability for Windows.

Switch 2.0.7

This release includes a large number of new features and compatibility and stability improvements. It should have minimal effect on results from existing models. Huge thanks are due to Josiah Johnston for most of these changes and to new contributors Desmond Zhong, Brad Venner and Martin Staadecker.

The updates are summarized below. For more details, see the git commit log.

Changes that may affect existing models and results

  • Changed Pyomo dependencies to versions 5.7.0-6.4.2. This has a few implications:
    • All previous users of Switch will need to upgrade their Pyomo installation, since Switch 2.0.6 was only compatible up through Pyomo 5.6.8. The upgrade to 5.7.0 is needed to get around a bug in Pyomo 5.6.8 and earlier that would incorrectly treat invalid optional data as missing data. If you have invalid data in optional columns, you may now receive error messages instead of it being silently ignored.
    • Note that Pyomo versions before 5.7.3 don't work with CPLEX version 12.10 or later (see Pyomo/pyomo#1792). If using a recent version of CPLEX, you should ensure that you also use version 5.7.3 or later of Pyomo.
    • If you have older models or custom modules, you should make the following changes to ensure compatibility with more recent versions of Pyomo:
      • assign dimen values for all Pyomo Sets.
      • initialize Pyomo Sets using ordered containers, not Python sets. (switch_model.utilities.unique_list can be useful for this)
      • explicitly specify a within domain for all Pyomo Sets and Params
      • don't use +inf, -inf, +infinity, -infinity or nan in input files (in some cases, one of these may be the default value for a parameter, so the input value can be specified as . instead)
  • The generation_projects_info.csv input file has been renamed to gen_info.csv. The new name is more consistent with other file names, which generally start with "gen_", and may be easier to remember and view in editor tabs. Switch 2.0.7 will offer to automatically update this file (and others) the first time it is run with an version 2.0.6 inputs directory.
  • The gen_predetermined_cap column has been renamed to build_gen_predetermined in gen_build_predetermined.csv. This is more consistent with the new build_gen_energy_predetermined column, and the use of 'build' in the names makes it clearer that these columns show the amount of new capacity built in each year, not the total amount of capacity in place that year.
  • The distribution_loss_rate column has been moved from the trans_params.csv input file to load_zones.csv and renamed local_td_loss_rate. This makes the name more accurate and allows it to be varied between load_zones.
  • The gen_multiple_fuels.dat input file has been replaced with gen_multiple_fuels.csv. Users should update their model setup scripts to create the new file. The .csv file should have two columns: GENERATION_PROJECT and fuel. It should have one row for each allowed fuel for each multi-fuel generator.
  • The dispatch-wide.csv output file has been renamed dispatch_wide.csv.
  • An extra "m" has been removed from the names of the cumulative_capacity_by_tech_periods.csv and cumulative_transmission_by_path_periods.csv output files produced by the reporting.basic_exports module.
  • Switch now implements a check for bidirectional transmission lines being specified in input files, so the implementation matches documentation.
  • Switch no longer uses the auto_select/autoselect argument in calls to switch_data.load_aug. Columns are always auto-selected now unless a 'select' argument is passed. You should remove this argument from any custom modules you use.
  • Curtailable storage charging is now included as a form of reserves in the spinning_reserves module, matching the spinning_reserves_advanced module. This may change the results from models that use storage with the spinning_reserve module.
  • balancing.demand_response_simple now defaults to providing 'spinning' reserves if a reserve module is loaded, otherwise providing no reserves.
  • balancing.demand_response.iterative now works with spinning_reserves_advanced, including a --demand-response-reserve-types flag. It also now reports final results in a file with no iteration number at the end.
  • The internal zone_rfm component has been renamed to zone_fuel_rfm to make it more clear that it identifies the rfm for each zone-fuel combination.
  • The output file gen_cap.csv now reports storage energy capacity (MWh) in place during each study period, in addition to power capacity (MW) in place. It also includes the capital recovery for the storage MWh in the reported annual capital recovery (previously omitted). The GenCapitalCosts column in this file has been renamed to GenCapitalRecovery to clarify that it is the amortized capital repayment that occurs for each project during each year, and distinguish it from capital outlay that occurs when the project is first built and is later recovered over time via the GenCapitalRecovery.
  • The new gen_build.csv output file shows the amount of power (MW) and storage (MWh) capacity added during each period. This also shows the total capital outlay needed for these additions. Switch uses capital outlay to calculate annual capital recovery requirements, then includes capital recovery each period in the system cost that it optimizes. Switch doesn't use the capital outlay for anything except calculating the annual capital recovery required. However, capital outlay may be of interest to planners.
  • Switch no longer produces the storage_builds.csv output file, since the information from this file is now shown in gen_build.csv and gen_cap.csv.
  • When using the balancing.unserved_load module and the transmission.local_td module, unserved load is now applied at the distribution node instead of the zone backbone node. This clarifies supply-demand balance and avoids some cases where reducing the load at the backbone may not be enough to avoid infeasibility. This may slightly change the unserved load and total cost reported in those models, due to avoiding the need for local transmission and distribution to meet the unserved portion of load.
  • Switch now gives a more compact description of the location of errors when they occur. If you would like to see the full Python traceback (the former behavior), use the new --full-traceback flag.
  • The balancing.planning_reserves module previously assigned a default capacity value for solar plants that could not exceed 1, even in timepoints when the capacity factor was greater than 1.0. We now allow values greater than 1 when calculating the default value for this parameter.
  • The --sorted-output flag is applied more universally. This ensures most outputs will be sorted when requested, including dispatch.csv, dispatch_wide.csv (columns), gen_cap.csv, gen_project_annual_summary.csv, load_balance.csv, local_td_energy_balance.csv, local_td_energy_balance_wide.csv and transmission.csv. Previous versions did not use this flag so universally, so you may find the order of some outputs changes when you upgrade.

New features

  • new --log-level flag can be set to error, warning, info or debug to receive varying amounts of information about the model run. --verbose is a now a synonym for --log-level info, and --quiet is equivalent to --log-level warning (default). Definitions of the levels are as follows:
    • error: may be used to give extra explanation when an exception is raised
    • warning: warn user about behavior that is most likely wrong but not enough to cause an exception (default output level, so users will see this for most models)
    • info: high-level progress log; used to follow progress of the model without seeing every detail
    • debug: detailed diagnostic data (e.g., recommend improved practices for input data files even if the current files are officially acceptable)
  • A new module balancing.diagnose_infeasibility can be used to help diagnose infeasible models, generally caused by inconsistent input data. This module relaxes all constraints, making any model feasible, then seeks to minimize all constraint violations, then reports which constraints are violated. Users can use a --no-relax CONSTRAINT1 [CONSTRAINT2 ...] argument to selectively turn some constraints back on, to identify combinations of constraints that cannot be enforced simultaneously. This is similar to CPLEX's irreducibly infeasible set (IIS) feature, but faster and simpler and works with all solvers.
  • Energy balance of the local T&D node is now exported to local_td_energy_balance.csv and local_td_energy_balance_wide.csv.
  • Updated simple_hydro to model spillways, allowing river flow to exceed the capacity of the generator & reservoir.
  • Storage decisions are now shown in dispatch-related output files if the storage module is included.
  • Expanded export from the generator dispatch module to simplify analysis. gen_project_annual_summary.csv shows energy production, emissions, capacity online, capital and O&M costs, levelized cost of energy, capacity factor and storage utilization for each generation project for each study year. dispatch_zonal_annual_summary.csv shows the same information per technology per load zone. dispatch_annual_summary.csv shows the same information per technology, aggregated across the entire region.
  • Implemented predetermined energy capacity for storage, a parallel to predetermined power capacity. This is specified in MWh in a new build_gen_energy_predetermined column in gen_build_predetermined.csv. It should be left blank (".") for non-storage projects, or you may omit the column if your model has no storage projects.
  • Added new command-line option --input-alias or --input-aliases. This allows users to specify replacements for individual input files to use in the current run. e.g., --input-alias fuel_cost.csv=fuel_cost_high.csv would use the fuel_cost_high.csv file instead of fuel_cost.csv in the current run. This can be used with one or more substitutions and can be repeated: --input-alias[es] file1.csv=file1.alternate.csv [file2.csv=file2.alternate.csv, ...]. These are applied as a simple replacement on the filename, then added onto the directory specified with --inputs-dir. So usually the replacement will occur within the --inputs-dir, but users can specify some_dir/file.csv or ../some_other_dir/file.csv in the alias to refer to files in other directories relative to the normal location of the file. Filename none will cause the file to be ignored.
  • When using the hawaii.ev module, users can now split the EV fleet between different charging modes using the --ev-timing flag: --ev-timing EV_TIMING [EV_TIMING ...]. Each EV_TIMING entry consists of a mode and optionally a share, e.g., --ev-timing bau charges all vehicles in bau mode or --ev-timing bau=0.32 optimal=0.68 splits the fleet between bau and optimal mode. Modes are bau=business-as-usual (upon arrival), flat=around the clock, or optimal (default). If modes are specified without shares assigned, they will receive equal fractions of the unallocated charging.
  • It is now possible to put comments on the same line as data in modules.txt, options.txt and scenarios.txt. This can be done by placing a # followed by the comment at the end of the line.
  • Zero-weight timeseries are now allowed. These are useful for modeling rare, worst-case days or including worse-than-worst (non-real) days as a form of planning reserves.
  • It is now possible to assign different fuel costs for each timepoint by using energy_sources.fuel_costs.simple_per_timepoint instead of energy_sources.fuel_costs.simple. It is also possible to set prices on different timescales (e.g., per timeseries) by assigning the same price to all timepoints in that period. This file reads input file fuel_cost_per_timepoint.csv, which should have these columns: fuel_cost_per_timepoint.csv load_zone, fuel, period, fuel_cost_per_timepoint.
  • The new fuel_costs.markets_expansion module allows capacity expansion of fuel markets. If using this module, you should add columns rfm_supply_tier_fixed_cost and rfm_supply_tier_max_age to the fuel_supply_curves.csv input file. These can be specified as . for tiers that are already available. For candidate tiers that may or may not be built, the rfm_supply_tier_fixed_cost specifies the fixed cost per MMBtu of fuel supply made available by that tier (not necessarily used) and rfm_supply_tier_max_age specifies the life of the tier if it is activated. This module is useful for considering investments in infrastructure that will expand the availability of fuels, such as a liquified natural gas (LNG) terminal. By testing tiers with different lives (and therefore different costs per MMBtu of fuel made available), it is possible to assess questions such as whether early fuel infrastructure investment will crowd out later renewable deployment. If the power system being studied will block use of a fuel after a certain date, then rfm_supply_tier_max_age should end before that date, to ensure that stranded costs after the end of the study are not omitted from the study. To do this, it may be necessary to add side constraints to prevent use of long-lived tiers after certain dates. The No_LNG_In_100_RPS constraint in switch_model.hawaii.lng_conversion shows an example of this.
  • The new transmission.copperplate module can be used to enable unlimited power transfer between zones at no cost.
  • Switch now allows general-purpose models with no zonal power demand and no power system components. This can be useful, e.g., for studying gas networks independently from the electricity system.
  • Progress in constructing model components is reported in 10% steps if --log-level is info. If --log-level is debug, these steps are shown along with timing to construct each individual component. Constructing model components is one of the lengthiest steps in Switch, so this gives more reassurance that something is happening.
  • New command line option --skip-generic-output tells switch_model.reporting not to save data for each model variable after the plan is optimized. These files can also be excluded by omitting the switch_model.reporting module from your model, which will also prevent creation of total_cost.txt and cost_components.csv. Specifying --no-post-solve will prevent running the post_solve code in all the modules used in the model, and therefore prevent all output.
  • Model configuration is now saved in model_config.json in outputs directory.
  • The demand_response.iterative module now includes the iteration number in result filenames, which can be useful for monitoring progress of the solution.
  • existing_local_td and local_td_annual_cost_per_mw are now optional (default 0) when using the local_td module. This is useful for models where existing local transmission and distribution capacity is unknown (Switch will automatically build enough) or where the cost of local T&D capacity is not important, e.g., if the user is only concerned about avoiding local T&D losses.
  • Switch will now report dual values (shadow prices) for the carbon cap in emissions.csv even for models with integer or binary variables. Note that dual values are not defined in general for integer models, and for this reason many solvers do not provide them. However, it is common practice to solve once, then fix integer variables at their current level, then re-solve the now continuous model and obtain dual values for that (this is what the --retrieve-cplex-mip-duals flag does with the cplex solver, and some versions of the cplexamp solver do this automatically). In this case, the shadow price of carbon assumes none of the integer variables move from their optimized position (e.g., not turning on one more power plant or abandoning the plan to build a plant with a specific minimum size), which should be suitable for most applications.
  • A solved model will be returned at the end of switch_model.solve.main() if the return_model and return_instance arguments are not specified. This can be useful for programmatic control of Switch.

Other changes

  • Switch code has begun to use m.logger.error(), m.logger.warning(), m.logger.info() and m.logger.debug() for screen output, and this is recommended for custom user modules too.
  • Changed plotting library from ggplot to plotnine for optional plots stored in dispatch_annual_summary_fuel.pdf and dispatch_annual_summary_tech.pdf (these plots will be generated automatically if you have plotnine installed in your Python environment)
  • Numerous adjustments to improve performance, documentation, warning messages or stability. See git commit log for details.
  • A bug was fixed in the balancing.planning_reserves module that may sometimes have caused Switch to crash.
  • Updated planning reserves input documentation & reading to reflect that some parameters are optional and don't have to be specified in files.
  • Fixed a bug with fuel unavailability when calculating fuel costs related to dispatch with the fuel_costs.markets module.
  • Switch now uses SwitchAbstractModel and SwitchConcreteModel classes to encapsulate the features we add to the Pyomo base classes. (Mainly of interest to core developers.)
  • Switch source code now uses the Black autoformatter. This should make the source code a little more readable, and is recommended for any contributions from users.

Switch 2.0.6

This release fixes a bug where the fixed costs of storage energy capacity (the MWh part of storage) from all possible build years were mistakenly applied each period, instead of only using the build years that are still in service in the current period. This increased the apparent cost of storage by approximately (study length) / (storage life). This bug was introduced in version 2.0.0b3 and persisted through version 2.0.5, so results from earlier models will need to be updated.

This will be the last version of Switch to work in Python 2. It requires at least Python 2.7.12 and also works with Python 3.

Switch 2.0.5

This release standardizes all inputs and outputs as .csv files.

As usual, when you first solve an older model, Switch will prompt to backup and upgrade the inputs directory. If you accept, it will convert the existing tab-delimited *.tab files and most ampl-format *.dat files to comma-delimited *.csv files. It is recommended that you update your model data preparation scripts to create .csv files directly. Note that non-indexed parameters should now be stored in .csv files with a header row listing the parameter names and a single data row showing their values.

All multi-value outputs from Switch are also now in comma-delimited .csv files, instead of a mix of .csv, .tab and .txt files. (total_cost.txt is unchanged)

This release also includes includes the following minor updates:

  • Updated installation instructions
  • Switch version number and website are now shown in the startup banner when running with --verbose flag; solve messages have also been improved slightly
  • Some parsing errors for *.tab files have been fixed in the upgrade scripts; this may cause errors during the upgrade process for input files that use spaces instead of tabs and were previously upgraded by Switch, producing malformed files.
  • Fixed several bugs in the documentation and execution of the stochastic examples that use the PySP module of the Pyomo package

Switch 2.0.4

This release introduces compatibility with Python 3. As of version 2.0.4, Switch can now be run with either Python 2.7 or Python 3 (likely to work with 2.7.10+; has been tested on 2.7.16 and 3.7.3).

This release will prompt to upgrade your model inputs directory, but the only change it makes is to update switch_inputs_version.txt to 2.0.4.

This release includes the following updates:

  • Code has been updated in many places to achieve Python 2/3 cross-compatibility. Future contributors should ensure that their code is compatible with both Python 2 and 3 (e.g., use switch_model.utilities.iteritems(dict) instead of dict.iteritems(), be prepared for results from dict.keys(), dict.vars(), map(), range(), zip(), etc., to be either generators or lists, and use from __future__ import division whenever doing division).
  • Installation instructions in INSTALL have been updated. We now recommend that users install dependencies using the conda command first, then install Switch using pip. This follows practices recommended in https://www.anaconda.com/using-pip-in-a-conda-environment/ and should minimize problems caused by incompatibilities between conda and pip.
  • Output files (.csv, .tab, .tsv, and .txt) are now consistently written using the local system's line endings (LF on Mac or Linux, CRLF on Windows). Previously, most of these were written with only LF line endings on Windows.
  • A bug was fixed in switch_model.transmission.local_td that prevented the carrying cost of Legacy local T&D capacity from being included in the objective function. As a result, users of this module will find that Switch now reports higher total costs than previously. However, this should not affect any of the decisions that Switch makes.
  • To make switch_model.transmission.local_td module compatible with Python 3, "Legacy" was removed from the list of build years for local T&D capacity (Pyomo sorts index keys when solving the model, and Python 3 cannot sort lists that mix strings and numbers). Legacy capacity is now read directly from the existing_local_td[z] parameter when needed. This does not change the behavior of Switch, but "Legacy" rows are no longer written to the BuildLocalTD.tab output file. The LOCAL_TD_BLD_YRS set has also been removed. LOAD_ZONES * PERIODS can be used instead.
  • A new indexed set, CURRENT_AND_PRIOR_PERIODS_FOR_PERIOD[p] has been added. This is useful for simple online capacity calculations for assets that cannot be retired during the study (e.g., AssetCapacity[p] = sum(BuildCapacity[v] for v in CURRENT_AND_PRIOR_PERIODS_FOR_PERIOD[p]))
  • Code has been cleaned up a bit internally (e.g., removed trailing whitespace, changed "SWITCH" or "SWITCH-Pyomo" to "Switch")

Switch 2.0.3

  • Users can now provide data in variable_capacity_factors.tab and hydro_timeseries.tab for times before projects are built or after they are retired without raising an error. However, the extra datapoints will be ignored.
  • Various parts of the code have better formatting, documentation and performance.
  • switch_model.hawaii.smooth_dispatch is now compatible with Pyomo 5.6 and later.
  • A new '--exact' option in switch_model.hawaii.rps forces the system to exactly meet the RPS target and no more. This is useful for studying the cost of adopting various levels of renewable power, including levels below the least-cost system design (i.e., cases where low shares of renewable power cause higher system costs).
  • A bug was fixed when calculating the cost of water spillage in switch_model.generators.extensions.hydro_system.
  • Final reservoir level in switch_model.generators.extensions.hydro_system is now stored in a varaible called ReservoirFinalVol. The ReservoirSurplus variable has been eliminated.
  • Bounds on a number of inputs have been relaxed to allow unusual or edge cases. In particular, a number of variables can now be zero instead of strictly positive. This allows zero costs, zero capacity limits, zero-based year counting, etc.
  • The gen_is_baseload parameter is now optional, with a default value of False (0).
  • NEW_TRANS_BLD_YRS has been renamed to TRANS_BLD_YRS.
  • setup.py now lists an optional dependency on rpy2<3.9 instead of rpy2, because later versions of rpy2 require Python 3, which Switch doesn't support yet. This only affects the iterative demand response module.
  • A new GENS_BY_ENERGY_SOURCE set can be used to identify all the generators that use any energy source, either a fuel or a non-fuel energy source. GENS_BY_FUEL and GENS_BY_NON_FUEL_ENERGY_SOURCE also still exist.
  • We have begun migrating toward using initialize instead of rule when initializing Pyomo components, and recommend that users do the same in their custom modules. This matches the current Pyomo API documentation. rule also works for now, but initialize should be more future proof.
  • The discrete-build requirement is now enforced on generators with predetermined build quantities, in addition to optimized generators.
  • The optional psycopg2 dependency has been changed to psycopg2-binary.
  • The --debug option now uses the ipdb debugger if available; otherwise it falls back to pdb.

Switch 2.0.2

  • General

    • Added --assign-current-version argument to switch upgrade. This is useful for updating version number in example directories to match current version of Switch, even if the data files don't need an upgrade.

  • Hawaii regional package

    • Fixed bug in hawaii.rps that would crash switch solve --help.

Switch 2.0.1

  • General

    • Switch is now compatible with Pyomo 5.6+ (in addition to earlier versions).
    • A new --no-post-solve option prevents all post-solve actions (e.g., saving variable results).
    • If the user specifies --reload-prior-solution, Switch now behaves as if it had just solved the model, i.e., after loading the solution, it runs post- solve code unless --no-post-solve is specified (useful for re-running reporting code), and it only drops to an interactive Python prompt if the user also specifies --interact.
    • A new --no-save-solution disables automatic solution-saving. This saves time and disk space for models that won't need to be reloaded later.
    • New --quiet and --no-stream-solver arguments cancel --verbose and --stream-solver.
    • A new "--save-expression[s] ..." argument can be used to save values for any Pyomo Expression object to a .tab file after the model is solved (similar to the automatic saving of variable values). This also works for Param objects.
    • The --include-module(s), --exclude-module(s), --save-expression(s), --suffix(es) and --scenario(s) flags can now be used repeatedly on the command line, in options.txt or in scenarios.txt. The include and exclude options will be applied in the order they are encountered, in options.txt, then scenarios.txt, then the command line.
    • A new --retrieve-cplex-mip-duals flag can be used to support retrieving duals for a MIP program from the cplex solver (users must also turn on the "duals") suffix. This flag patches the Pyomo-generated cplex command script to pass the "change problem fix" command to the solver and then solve a second time. This fixes integer variables at their final values, then re-solves to obtain duals. This flag is not needed with the cplexamp solver.
    • A new balancing.demand_response.iterative module has been added. This was formerly in the Hawaii regional package. This module performs iterative solutions with any convex demand system, based on a bid-response process.
    • New indexed sets have been added to allow efficient selection of groups of generators that use a particular technology, fuel or non-fuel energy source: GENS_BY_TECHNOLOGY, GENS_BY_FUEL, GENS_BY_NON_FUEL_ENERGY_SOURCE.
    • Generator capacity data is saved to gen_cap.tab instead of gen_cap.txt and rows are sorted if user specifies --sorted-output.
    • Even if a model has solver warnings, results will be reported and post-solve will be performed if a valid solution is available.
    • A more descriptive warning is given when switch_model.reporting finds an uninitialized variable.
    • A warning is given about potential errors parsing arguments in the form "--flag=value". Python's argument parsing module can make mistakes with these, so "--flag value" is a safer choice.
    • Switch now monkeypatches Pyomo to accelerate reloading prior solutions. Previously Pyomo 5.1.1 (and maybe others) took longer to load prior solutions than solving the model.
    • At startup, "switch solve-scenarios" will restart jobs that were previously interrupted after being started by the same worker (same --job-id argument or SWITCH_JOB_ID environment variable). Note that workers automatically pull scenarios from the scenario list file until there are none left to solve, and avoid solving scenarios that have been pulled by other workers. Each worker should be given a unique job ID, and this ID should be reused if the worker is terminated and restarted. The new behavior ensures that jobs are not abandoned if a worker is restarted.

  • Upgrade scripts

    • The upgrade scripts now report changes in module behavior or renamed modules while upgrading an inputs directory. This only reports changes to modules used in the current model.
    • The hawaii.reserves module is automatically replaced by balancing.operating_reserves.areas and balancing.operating_reserves.spinning_reserves in the module list.
    • The hawaii.demand_response module is replaced by balancing.demand_response.iterative and hawaii.r_demand_system is replaced by balancing.demand_response.iterative.r_demand_system in the module list.
    • "switch_mod" will not be changed to "switch_modelel" if a module file is upgraded from 2.0.0b1 to 2.0.0b2 twice.

  • Hawaii regional package

    • The hawaii.reserves module has been deprecated and the hawaii.demand_response module has been moved (see upgrade scripts)
    • Switch now places limits on down-reserves from pumped-storage hydro.
    • A new --rps-prefer-dist-pv option for hawaii.rps will prevent construction of new large PV until 90% of distributed PV potential has been developed.
    • Limits on load-shifting between hours in hawaii.demand_response_simple have been formalized.
    • The Hawaii storage energy cost calculation has been fixed.
    • Total production by energy source is reported by hawaii.save_results, ad-hoc technologies are added to production_by_technology.tsv, and hourly dispatch is disaggregated by non-fuel technologies.
    • Bugs have been fixed in reserve calculation for EVs and in hawaii.smooth_dispatch and hawaii.scenario_data.
    • hawaii.smooth_dispatch minimizes total inter-hour change instead of square of levels. The old quadratic smoothing method has been moved to hawaii.smooth_dispatch.quadratic (slow and possibly buggy).
    • The must-run requirement in hawaii.kalaeloa is turned off when RPS or EV share is above 75% (can be overridden by --run-kalaeloa-even-with-high-rps)
    • Support for nominal-dollar fuel price forecasts has been dropped from hawaii.scenario_data
    • A new --no-hydrogen flag can be used to deactivate the hydrogen module.
    • The hawaii.ev_advanced module now calculates vehicle fleet emissions.

Switch 2.0.0

First public release of Switch 2. This uses a similar framework to Switch 1, but with numerous improvements. The most significant are:

  • Written in Python instead of AMPL language
  • Modular approach, so components can be easily added or removed from model
  • Modeling of unit commitment and part load heat rates (optional)
  • Generalization of sample timeseries to have arbitrary length instead of single days
  • Standardized reporting, e.g., automatic export of all variable values