.. _demand_saving:

Demand saving
-------------

Demand saving is a common concept in water resource models. When the state of a water resource is poor (for example, lower than normal reservoir levels) demand restrictions may be imposed on customers in order to reduce the likelihood of a failure to supply.

The demand at a given timestep can be represented using the equation below,

.. math::

    d = d_{mean} * p * (1-s)

where :math:`d` is the baseline demand to be calculated, :math:`d_{mean}` is the baseline demand, :math:`p` is the annual profile and :math:`s` is the demand saving (e.g. 5%).

The code snippets that follow belong in the ``"parameters": {}`` section of the model document (except for the ``node`` at the end).

The baseline demand is specified as a constant (this is often the *mean* demand).

.. code-block:: javascript

    "demand_baseline": {
        "type": "constant",
        "values": 50
    }

A daily or monthly profile can be used to vary the demand throughout the year. In the example below the demand in May - August is 1.2x the baseline demand, with the rest of the year at 0.9x the baseline, forming the common "top hat" profile (illustrated below).

.. code-block:: javascript

    "demand_profile": {
        "type": "monthlyprofile",
        "values": [0.9, 0.9, 0.9, 0.9, 1.2, 1.2, 1.2, 1.2, 0.9, 0.9, 0.9, 0.9]
    },

.. plot:: pyplots/top_hat.py

The demand saving level describes the level of restrictions as an integer, where 0 means no demand restrictions, level 1 (L1) is some restriction (or perhaps just publicity), level 2 (L2) more severe restrictions, and so on. The specifics will depend on the system being modelled. This is represented in Pywr using the ``ControlCurveIndexParameter``. ``IndexParameter`` return an integer to be used as an index, rather than a decimal value.

The demand saving level is often related to the storage of strategic reservoirs in relation to a control curve (the expected reservoir volume at a given time of the year). The example below two demand saving levels are defined (and implicitly a L0) based on the volume in the ``"Central Reservoir"`` storage node. L1 is defined as 80% of full, L2 as 50% of full. Control curves are commonly more complicated, as the expected level of a reservoir is usually lower in the summer than it is in the winter.

.. code-block:: javascript

    "demand_saving_level": {
        "type": "controlcurveindex",
        "storage_node": "Central Reservoir",
        "control_curves": [
            "level1",
            "level2"
        ]
    },
    "level1": {
        "type": "constant",
        "values": 0.8
    },
    "level2": {
        "type": "constant",
        "values": 0.5
    },

The demand saving factor is calculated by indexing an array of demand saving profiles with the demand saving level. In the example below the list of profiles (``"params"``) corresponds to the L0, L1 and L2 profiles respectively. At L0 a constant ``1.0`` is used to represent no savings. At L1 there is a 10% reduction in demand (``0.90`` as a factor) during the summer months and a 5% reduction elsewhere (``0.95``). At L2 there are further reductions to 75%/80%.

.. code-block:: javascript

    "demand_saving_factor": {
        "type": "indexedarray",
        "index_parameter": "demand_saving_level",
        "params": [
            {
                "type": "constant",
                "values": 1.0
            },
            {
                "type": "monthlyprofile",
                "values": [0.95, 0.95, 0.95, 0.95, 0.90, 0.90, 0.90, 0.90, 0.95, 0.95, 0.95, 0.95]
            },
            {
                "type": "monthlyprofile",
                "values": [0.8, 0.8, 0.8, 0.8, 0.75, 0.75, 0.75, 0.75, 0.8, 0.8, 0.8, 0.8]
            }
        ]
    },

To understand how the index works, the following equivalent Python code may help:

.. code-block:: python

    month = 5
    demand_saving_level = 1
    demand_factors = [[1.0, ...], [0.95, ...], [0.8, ...]]
    demand_saving_factor = demand_factors[demand_saving_level][month-1]

Finally the demand components can be combined as in the equation at the beginning using an ``AggregatedParameter``. Each timestep the value of each of the components is calculated and the values are multiplied to give the final demand value.

.. code-block:: javascript

    "demand_max_flow": {
        "type": "aggregated",
        "agg_func": "product",
        "parameters": [
            "demand_baseline",
            "demand_profile",
            "demand_saving_factor"
        ]
    },

The final profiles are illustrated in the figure below. The actual demand value will switch between the three profiles depending on the resource state of the reservoir.

.. plot:: pyplots/demand_saving_levels.py

This parameter can then be applied to the ``max_flow`` attribute of the demand node.

.. code-block:: javascript

    {
        "type": "output",
        "name": "Demand",
        "max_flow": "demand_max_flow",
        "cost": -500
    },

When a model has more than one demand node it is OK to re-use the demand saving level/factor for each demand node.