Advanced features

Once you’re comfortable with building, running, and analysing one of the built-in example models, you may want to explore Calliope’s advanced functionality. With these features, you will be able to build and run complex models in no time.

Time resolution adjustment

Models have a default timestep length (defined implicitly by the timesteps of the model’s time series data). This default resolution can be adjusted over parts of the dataset by specifying time resolution adjustment in the model configuration, for example:

        function: resample
        function_options: {'resolution': '6H'}

In the above example, this would resample all time series data to 6-hourly timesteps.

Calliope’s time resolution adjustment functionality allows running a function that can perform arbitrary adjustments to the time series data in the model.

The available options include:

  1. Uniform time resolution reduction through the resample function, which takes a pandas-compatible rule describing the target resolution (see above example).

  2. Deriving representative days from the input time series, by applying the clustering method implemented in calliope.time.clustering, for example:

        function: apply_clustering
            clustering_func: kmeans
            how: mean
            k: 20

When using representative days, a number of additional constraints are added, based on the study undertaken by Kotzur et al. These constraints require a new decision variable storage_inter_cluster, which tracks storage between all the dates of the original timeseries. This particular functionality can be disabled by including storage_inter_cluster: false in the function_options given above.


It is also possible to load user-defined representative days, by pointing to a file in clustering_func in the same format as pointing to timeseries files in constraints, e.g. clustering_func: file=clusters.csv:column_name. Clusters are unique per datestep, so the clustering file is most readable if the index is at datestep resolution. But, the clustering file index can be in timesteps (e.g. if sharing the same file as a constraint timeseries), with the cluster number repeated per timestep in a day. Cluster values should be integer, starting at zero.

  1. Heuristic selection of time steps, that is, the application of one or more of the masks defined in calliope.time.masks, which will mark areas of the time series to retain at maximum resolution (unmasked) and areas where resolution can be lowered (masked). Options can be passed to the masking functions by specifying options. A time.function can still be specified and will be applied to the masked areas (i.e. those areas of the time series not selected to remain at the maximum resolution), as in this example, which looks for the week of minimum and maximum potential wind generation (assuming a wind technology was specified), then reduces the rest of the input time series to 6-hourly resolution:

            - {function: extreme, options: {padding: 'calendar_week', tech: 'wind', how: 'max'}}
            - {function: extreme, options: {padding: 'calendar_week', tech: 'wind', how: 'min'}}
        function: resample
        function_options: {'resolution': '6H'}


When using time clustering or time masking, the resulting timesteps will be assigned different weights depending on how long a period of time they represent. Weights are used for example to give appropriate weight to the operational costs of aggregated typical days in comparison to individual extreme days, if both exist in the same processed time series. The weighting is accessible in the model data, e.g. through Model.inputs.timestep_weights. The interpretation of results when weights are not 1 for all timesteps requires caution. Production values are not scaled according to weights, but costs are multiplied by weight, in order to weight different timesteps appropriately in the objective function. This means that costs and production values are not consistent without manually post-processing them by either multipyling production by weight (production would then be inconsistent with capacity) or dividing costs by weight. The computation of levelised costs and of capacity factors takes weighting into account, so these values are consisten and can be used as usual.

See also

See the implementation of constraints in calliope.backend.pyomo.constraints for more detail on timestep weights and how they affect model constraints.

Setting a random seed

By specifying model.random_seed in the model configuration, any alphanumeric string can be used to initialise the random number generator at the very start of model processing.

This is useful for full reproducibility of model results where time series clustering is used, as clustering methods such as k-means depend on randomly generated initial conditions.

Note that this affects only the random number generator used in Calliope’s model preprocessing and not in any way the solver used to solve the model (any solver-specific options need to be set specifically for that solver; see Specifying custom solver options).

Using tech_groups to group configuration

In a large model, several very similar technologies may exist, for example, different kinds of PV technologies with slightly different cost data or with different potentials at different model locations.

To make it easier to specify closely related technologies, tech_groups can be used to specify configuration shared between multiple technologies. The technologies then give the tech_group as their parent, rather than one of the abstract base technologies.

You can as well extend abstract base technologies, by adding an attribute that will be in effect for all technologies derived from the base technology. To do so, use the name of the abstract base technology for your group, but omit the parent.

For example:

                interest_rate: 0.1
            parent: supply
            carrier: power
            resource: file=pv_resource.csv
            lifetime: 30
                om_annual_investment_fraction: 0.05
                depreciation_rate: 0.15

            parent: pv
            name: 'Large-scale PV'
            energy_cap_max: 2000
                energy_cap: 750
            parent: pv
            name: 'Rooftop PV'
            energy_cap_max: 10000
                energy_cap: 1000

None of the tech_groups appear in model results, they are only used to group model configuration values.

Operational mode

In planning mode, constraints are given as upper and lower boundaries and the model decides on an optimal system configuration. In operational mode, all capacity constraints are fixed and the system is operated with a receding horizon control algorithm.

To specify a runnable operational model, capacities for all technologies at all locations must have be defined. This can be done by specifying energy_cap_equals. In the absence of energy_cap_equals, constraints given as energy_cap_max are assumed to be fixed in operational mode.

Operational mode runs a model with a receding horizon control algorithm. This requires two additional settings:

        horizon: 48  # hours
        window: 24  # hours

horizon specifies how far into the future the control algorithm optimises in each iteration. window specifies how many of the hours within horizon are actually used. In the above example, decisions on how to operate for each 24-hour window are made by optimising over 48-hour horizons (i.e., the second half of each optimisation run is discarded). For this reason, horizon must always be larger than window.

Generating scripts to run a model many times

Scenarios and overrides can be used to run a given model multiple times with slightly changed settings or constraints.

This functionality can be used together with the calliope generate_runs and calliope generate_scenarios command-line tools to generate scripts that run a model many times over in a fully automated way, for example, to explore the effect of different technology costs on model results.

calliope generate_runs, at a minimum, must be given the following arguments:

  • the model configuration file to use

  • the name of the script to create

  • --kind: Currently, three options are available. windows creates a Windows batch (.bat) script that runs all models sequentially, bash creates an equivalent script to run on Linux or macOS, bsub creates a submission script for a LSF-based high-performance cluster, and sbatch creates a submission script for a SLURM-based high-performance cluster.

  • --scenarios: A semicolon-separated list of scenarios (or overrides/combinations of overrides) to generate scripts for, for example, scenario1;scenario2 or override1,override2a;override1,override2b. Note that when not using manually defined scenario names, a comma is used to group overrides together into a single model – in the above example, override1,override2a would be applied to the first run and override1,override2b be applied to the second run

A fully-formed command generating a Windows batch script to run a model four times with each of the scenarios “run1”, “run2”, “run3”, and “run4”:

calliope generate_runs model.yaml run_model.bat --kind=windows --scenarios "run1;run2;run3;run4"

Optional arguments are:

  • --cluster_threads: specifies the number of threads to request on a HPC cluster

  • --cluster_mem: specifies the memory to request on a HPC cluster

  • --cluster_time: specifies the run time to request on a HPC cluster

  • --additional_args: A text string of any additional arguments to pass directly through to calliope run in the generated scripts, for example, --additional_args="--debug".

  • --debug: Print additional debug information when running the run generation script.

An example generating a script to run on a bsub-type high-performance cluster, with additional arguments to specify the resources to request from the cluster:

calliope generate_runs model.yaml --kind=bsub --cluster_mem=1G --cluster_time=100 --cluster_threads=5  --scenarios "run1;run2;run3;run4"

Running this will create two files:

  • The cluster submission script to pass to bsub on the cluster.

  • The accompanying script defining the runs for the cluster to execute.

In all cases, results are saved into the same directory as the script, with filenames of the form out_{run_number}_{scenario_name}.nc (model results) and plots_{run_number}_{scenario_name}.html (HTML plots), where {run_number} is the run number and {scenario_name} is the name of the scenario (or the string defining the overrides applied). On a cluster, log files are saved to files with names starting with log_ in the same directory.

Finally, the calliope generate_scenarios tool can be used to quickly generate a file with scenarios definition for inclusion in a model, if a large enough number of overrides exist to make it tedious to manually combine them into scenarios. Assuming that in model.yaml a range of overrides exist that specify a subset of time for the years 2000 through 2010, called “y2000” through “y2010”, and a set of cost-related overrides called “cost_low”, “cost_medium” and “cost_high”, the following command would generate scenarios with combinations of all years and cost overrides, calling them “run_1”, “run_2”, and so on, and saving them to scenarios.yaml:

calliope generate_scenarios model.yaml scenarios.yaml y2000;y2001;y2002;2003;y2004;y2005;y2006;2007;2008;y2009;2010 cost_low;cost_medium;cost_high --scenario_name_prefix="run_"

Importing other YAML files in overrides

When using overrides (see Scenarios and overrides), it is possible to have import statements within overrides for more flexibility. The following example illustrates this:

            some_tech.constraints.energy_cap_max: 10
        import: [additional_definitions.yaml]


    some_other_tech.constraints.energy_eff: 0.1

This is equivalent to the following override:

            some_tech.constraints.energy_cap_max: 10
            some_other_tech.constraints.energy_eff: 0.1

Interfacing with the solver backend

On loading a model, there is no solver backend, only the input dataset. The backend is generated when a user calls run() on their model. Currently this will call back to Pyomo to build the model and send it off to the solver, given by the user in the run configuration run.solver. Once built, solved, and returned, the user has access to the results dataset model.results and interface functions with the backend model.backend.

You can use this interface to:

  1. Get the raw data on the inputs used in the optimisation.

    By running model.backend.get_input_params() a user get an xarray Dataset which will look very similar to model.inputs, except that assumed default values will be included. You may also spot a bug, where a value in model.inputs is different to the value returned by this function.

  2. Update a parameter value.

    If you are interested in updating a few values in the model, ou can run model.backend.update_param() . For example, to update your the energy efficiency of your ccgt technology in location region1 from 0.5 to 0.1, you can run model.backend.update_param('energy_eff', 'region1::ccgt`, 0.1). This will not affect results at this stage, you’ll need to rerun the backend (point 4) to optimise with these new values.

  3. Activate / Deactivate a constraint or objective.

    Constraints can be activated and deactivate such that they will or will not have an impact on the optimisation. All constraints are active by default, but you might like to remove, for example, a capacity constraint if you don’t want there to be a capacity limit for any technologies. Similarly, if you had multiple objectives, you could deactivate one and activate another. The result would be to have a different objective when rerunning the backend.


Currently Calliope does not allow you to build multiple objectives, you will need to understand Pyomo and add an additional objective yourself to make use of this functionality. The Pyomo ConcreteModel() object can be accessed at model._backend_model.

  1. Rerunning the backend.

    If you have edited parameters or constraint activation, you will need to rerun the optimisation to propagate the effects. By calling model.backend.rerun(), the optimisation will run again, with the updated backend. This will not affect your model, but instead will return a dataset of the inputs/results associated with that specific rerun. It is up to you to store this dataset as you see fit. model.results will remain to be the initial run, and can only be overwritten by


By calling any updates you have made to the backend will be overwritten.

Specifying custom solver options


Refer to the Gurobi manual, which contains a list of parameters. Simply use the names given in the documentation (e.g. “NumericFocus” to set the numerical focus value). For example:

    solver: gurobi
        Threads: 3
        NumericFocus: 2


Refer to the CPLEX parameter list. Use the “Interactive” parameter names, replacing any spaces with underscores (for example, the memory reduction switch is called “emphasis memory”, and thus becomes “emphasis_memory”). For example:

    solver: cplex
        mipgap: 0.01
        mip_polishafter_absmipgap: 0.1
        emphasis_mip: 1
        mip_cuts: 2
        mip_cuts_cliques: 3

Previous: Advanced constraints | Next: Configuration and defaults