Math syntax¶

This page provides an overview of the syntax available to formulate math components. reference for the allowed key-value pairs in your custom math YAML file is available in the reference section of the documentation.

foreach lists¶

If the math component is indexed over sets (e.g., techs, nodes, timesteps), then you need to define a foreach list of those sets. If the component is dimensionless, no foreach list needs to be defined.

For example, foreach: [nodes, techs] will build the component over all nodes and techs in the model.

The available sets in Calliope are: nodes, techs, carriers, costs, timesteps. If using time clustering and inter-cluster storage math, there is also a datesteps set available. If you want to build over your own custom set, you will need to add it to the Calliope model dataset before building the optimisation problem, e.g. as a new indexed parameter.

where strings¶

Where strings allow you to define math that applies to only a subset of your data or of the models you are running. They are made up of a series of statements combined with logical operators. These statements can be one of the following:

Checking the existence of set items in an input parameter. When checking the existence of an input parameter it is possible to first sum it over one or more of its dimensions; if at least one value on the summed dimension(s) is defined, then it will be considered defined in the remaining dimensions.
Examples
- If you want to apply a constraint across all nodes and techs, but only for node+tech combinations where the flow_out_eff parameter has been defined, you would include flow_out_eff.
- If you want to apply a constraint over techs and timesteps, but only for combinations where the source_use_max parameter has at least one node with a value defined, you would include any(resource, over=nodes). (1)
1. any is a helper function; read more below!
Checking the value of a configuration option or an input parameter. Checks can use any of the operators: >, <, =, <=, >=. Configuration options are any that are defined in config.build, where you can define your own options to access in the where string.
Examples
- If you want to apply a constraint only if the configuration option config.build.mode is operate, you would include config.mode=operate.
- If you want to apply a constraint across all nodes and techs, but only where the flow_eff parameter is less than 0.5, you would include flow_eff<0.5.
- If you want to apply a constraint only for the first timestep in your timeseries, you would include timesteps=get_val_at_index(dim=timesteps, idx=0). (1)
- If you want to apply a constraint only for the last timestep in your timeseries, you would include timesteps=get_val_at_index(dim=timesteps, idx=-1).
1. get_val_at_index is a helper function; read more below!
Checking the base_tech of a technology (storage, supply, etc.) or its inheritance chain (if using tech_groups and the inherit parameter).
Examples
- If you want to create a decision variable across only storage technologies, you would include base_tech=storage.
- If you want to apply a constraint across only your own rooftop_supply technologies (e.g., you have defined rooftop_supply in tech_groups and your technologies pv and solar_thermal define inherit: rooftop_supply), you would include inheritance(rooftop_supply). Note that base_tech=... is a simple check for the given value of base_tech, while inheritance() is a helper function (see below) which can deal with the fact that intermediate groups may be present, e.g. pv might inherit from rooftop_supply which in turn might inherit from electricity_supply.
Subsetting a set. The sets available to subset are always [nodes, techs, carriers] + any additional sets defined by you in foreach.
Examples
- If you want to filter nodes where any of a set of techs are defined: defined(techs=[tech1, tech2], within=nodes, how=any) (1).
1. defined is a helper function; read more below!

To combine statements you can use the operators and/or. You can also use the not operator to negate any of the statements. These operators are case insensitive, so "and", "And", "AND" are equivalent. You can group statements together using the () brackets. These statements will be combined first.

Examples

If you want to apply a constraint for storage technologies if the configuration option cyclic_storage is activated and it is the last timestep of the series: base_tech=storage and cyclic_storage=True and timesteps=get_val_at_index(dim=timesteps, idx=-1).
If you want to create a decision variable for the input carriers of conversion technologies: carrier_in and base_tech=conversion
If you want to apply a constraint if the parameter source_unit is energy_per_area or the parameter area_use_per_flow_cap is defined: source_unit=energy_per_area or area_use_per_flow_cap.
If you want to apply a constraint if the parameter flow_out_eff is less than or equal to 0.5 and source_use has been defined, or flow_out_eff is greater than 0.9 and source_use has not been defined: (flow_out_eff<=0.5 and source_use) or (flow_out_eff>0.9 and not source_use).

Combining foreach and where will create an n-dimensional boolean array. Wherever index items in this array are True, your component expression(s) will be applied.

expression strings¶

As with where strings, expression strings are a series of math terms combined with operators. The terms can be input parameters, decision variables, global expressions, or numeric values that you define on-the-fly.

If you are defining a global expression or objective, then the available expression string operators are: +, -, *, /, and ** ("to the power of"). These expressions are applied using standard operator precedence (BODMAS/PEMDAS, see this wiki for more info).

If you are defining a constraint, then you also need to define a comparison operator: <=, >=, or ==.

Examples

If you want to limit all technology outflow to be less than 200 units: flow_out <= 200.
If you want to create a global expression which is the storage level minus a parameter defining a minimum allowed storage level: storage - storage_cap * min_storage_level.
If you want to set the outflow of a specific technology my_tech to equal all outflows of a specific carrier my_carrier at each node: flow_out[techs=my_tech] == sum(flow_out[carriers=my_carrier], over=techs).
If you want inflows at a node my_node to be at least as much as the inflows in the previous timestep: flow_in[nodes=my_node] >= roll(flow_in[nodes=my_node], timesteps=1).

Slicing data¶

You do not need to define the sets of math components in expressions, unless you are actively "slicing" them. Behind the scenes, we will make sure that every relevant element of the defined foreach sets are matched together when applying the expression (we merge the underlying xarray DataArrays). Slicing math components involves appending the component with square brackets that contain the slices, e.g. flow_out[carriers=electricity, nodes=[A, B]] will slice the flow_out decision variable to focus on electricity in its carriers dimension and only has two nodes (A and B) on its nodes dimension. To find out what dimensions you can slice a component on, see your input data (model.inputs) for parameters and the definition for decision variables in your loaded math dictionary (model.math.variables).

Helper functions¶

For where strings and expression strings, there are many helper functions available to use, to allow for more complex operations to be undertaken. Their functionality is detailed in the helper function API page. Here, we give a brief summary. Some of these helper functions require a good understanding of their functionality to apply, so make sure you are comfortable with them before using them.

inheritance¶

using inheritance(...) in a where string allows you to grab a subset of technologies that all share the same tech_group in the technology's inherit key. If a tech_group also inherits from another tech_group (chained inheritance), you will get all techs that are children along that inheritance chain.

So, for the definition:

tech_groups:
  techgroup1:
    inherit: techgroup2
    flow_cap_max: 10
  techgroup2:
    base_tech: supply
techs:
  tech1:
    inherit: techgroup1
  tech2:
    inherit: techgroup2

inheritance(techgroup1) will give the [tech1] subset and inheritance(techgroup2) will give the [tech1, tech2] subset.

any¶

Parameters are indexed over multiple dimensions. Using any(..., over=...) in a where string allows you to check if there is at least one non-NaN value in a given dimension (akin to xarray.DataArray.any). So, any(cost, over=[nodes, techs]) will check if there is at least one non-NaN tech+node value in the costs dimension (the other dimension that the cost decision variable is indexed over).

defined¶

Similar to any, using defined(..., within=...) in a where string allows you to check for non-NaN values along dimensions. In the case of defined, you can check if e.g., certain technologies have been defined within the nodes or certain carriers are defined within a group of techs or nodes.

So, for the definition:

techs:
  tech1:
    base_tech: conversion
    carrier_in: electricity
    carrier_out: heat
  tech2:
    base_tech: conversion
    carrier_in: [coal, biofuel]
    carrier_out: electricity
nodes:
  node1:
    techs: {tech1}
  node2:
    techs: {tech1, tech2}

defined(carriers=electricity, within=techs) would yield a list of [True, True] as both technologies define electricity.

defined(techs=[tech1, tech2], within=nodes) would yield a list of [True, True] as both nodes define at least one of tech1 or tech2.

defined(techs=[tech1, tech2], within=nodes, how=all) would yield a list of [False, True] as only node2 defines both tech1 and tech2.

sum¶

Using sum(..., over=) in an expression allows you to sum over one or more dimension of your component array (be it a parameter, decision variable, or global expression).

select_from_lookup_arrays¶

Some of our arrays in model.inputs are not data arrays, but "lookup" arrays. These arrays are used to map the array's index items to other index items. For instance when using time clustering, the lookup_cluster_last_timestep array is used to get the timestep resolution and the stored energy for the last timestep in each cluster. Using select_from_lookup_arrays(..., dim_name=lookup_array) allows you to apply this lookup array to your data array.

get_val_at_index¶

If you want to access an integer index in your dimension, use get_val_at_index(dim_name=integer_index). For example, get_val_at_index(timesteps=0) will get the first timestep in your timeseries, get_val_at_index(timesteps=-1) will get the final timestep. This is mostly used when conditionally applying a different expression in the first / final timestep of the timeseries.

It can be used in the where string (e.g., timesteps=get_val_at_index(timesteps=0) to mask all other timesteps) and the expression string (via slices - storage[timesteps=$first_timestep] and first_timestep expression being get_val_at_index(timesteps=0)).

roll¶

We do not use for-loops in our math. This can be difficult to get your head around initially, but it means that to define expressions of the form var[t] == var[t-1] + param[t] requires shifting all the data in your component array by N places. Using roll(..., dimension_name=N) allows you to do this. For example, roll(storage, timesteps=1) will shift all the storage decision variable objects by one timestep in the array. Then, storage == roll(storage, timesteps=1) + 1 is equivalent to applying storage[t] == storage[t - 1] + 1 in a for-loop.

default_if_empty¶

We work with quite sparse arrays in our models. So, although your arrays are indexed over e.g., nodes, techs and carriers, a decision variable or parameter might only have one or two values in the array, with the rest being NaN. This can play havoc with defining math, with nan values making their way into your optimisation problem and then killing the solver or the solver interface. Using default_if_empty(..., default=...) in your expression string allows you to put a placeholder value in, which will be used if the math expression unavoidably needs a value. Usually you shouldn't need to use this, as your where string will mask those NaN values. But if you're having trouble setting up your math, it is a useful function to getting it over the line.

Note

Our internally defined parameters all have default values which propagate to the math. You only need to use default_if_empty for decision variables and global expressions, and for user-defined parameters.

equations¶

Equations are combinations of expression strings and where strings. You define one or more equations for your model components. A different where string associated with each equation expression allows you to slightly alter the expression for different component members. You define equations as lists of dictionaries:

equations:
  - where: ...
    expression: ...
  - where: ...
    expression: ...

If you are supplying only one equation, you do not need to define a where string:

equations:
  - expression: ...

Note

where strings within equations are appended to your top-level where string, e.g.:

where: storage_cap_max
equations:
  - where: flow_in_eff > 0.5  # <- this will be parsed as "storage_cap_max and flow_in_eff > 0.5"
  expression: ...
  - where: flow_in_eff <= 0.5  # <- this will be parsed as "storage_cap_max and flow_in_eff <= 0.5"
  expression: ...

Examples

Divide by efficiency if efficiency is larger than zero, otherwise set the variable to zero:

equations:
  - where: flow_eff > 0
    expression: flow_out / flow_out_eff == flow_in
  - where: flow_eff = 0
    expression: flow_out == 0

Limit flow by storage_cap, if it is defined, otherwise by flow_cap:

equations:
  - where: storage_cap
    expression: flow_out <= 0.5 * storage_cap
  - where: not storage_cap
    expression: flow_out <= 0.9 * flow_cap

Warning

You have to be careful when setting up different where strings to avoid clashes, where different expressions are valid for the same component member. We will raise errors when this happens, and if your where strings become too restrictive and so miss a component member that needs an expression.

sub-expressions¶

For long expressions - or those where a part of the expression might change for different component members according to a specific condition - you can define sub-expressions. These look similar to equations; they are lists of dictionaries with where and expression strings. They are accessed from you main expression(s) by reference to their name prepended with the special $ character. For example:

equations:
  - expression: flow_out <= $adjusted_flow_in
sub_expressions:
  adjusted_flow_in:
    - where: inheritance(storage)
      # main expression becomes `flow_out <= flow_in * flow_eff`
      expression: flow_in * flow_eff
    - where: inheritance(supply)
      # main expression becomes `flow_out <= flow_in * flow_eff * parasitic_eff`
      expression: flow_in * flow_eff * parasitic_eff
    - where: inheritance(conversion)
      # main expression becomes `flow_out <= flow_in * flow_eff * 0.3`
      expression: flow_in * flow_eff * 0.3

Note

As with equations, where strings are mixed in together. If you have two equation expressions and three sub-expressions, each with two expressions, you will end up with 2 * 3 * 2 = 12 unique where strings with linked expression strings.

slices¶

Similarly to sub-expressions, you can use references when slicing your data, again using the $ identifier. Standard slicing only allows for dimensions to reference plain strings or lists of plain strings. If you want to slice using a "lookup" parameter, you will need to provide it within the slices sub-key, e.g.:

If you define a lookup parameter "lookup_techs" as:

parameters:
  lookup_techs:
    data: True
    index: [tech_1, tech_2]
    dimensions: [techs]

Then the following slice will select only the tech_1 and tech_2 members of flow_out:

equations:
  - expression: sum(flow_out[carriers=electricity, techs=$tech_ref]) <= flow_in[carriers=heat] * 0.6
slices:
  tech_ref:
    - expression: lookup_techs