Change it and solve again¶

Ask the model from Your data what-if questions, and let the solver keep its model between them. Every block runs from the examples/ folder. We start from the sources of the last tutorial:

import polars as pl

import specsolve as sps

generators = pl.read_csv('tutorial/data/generators.csv')
load = pl.read_csv('tutorial/data/load.csv')
sources = {
    'generator': generators['generator'],
    'snapshot': load['snapshot'],
    'p_max': generators.select('generator', value='p_max'),
    'cost': generators.select('generator', value='cost'),
    'load': load,
}

Build once¶

build keeps the model that solve throws away, so you can solve it more than once:

model = sps.build('dispatch.yaml', sources)
print(model.solve().objective)

10500.0

The answer is 10500, as before.

New numbers¶

How much would more wind save? update puts new numbers on the model the solver holds. We try four wind capacities:

rows = []
for wind in (80.0, 120.0, 160.0, 200.0):
    p_max = pl.DataFrame({'generator': ['wind', 'solar', 'gas'], 'value': [wind, 0.0, 200.0]})
    rows.append({'wind': wind, 'objective': model.update({'p_max': p_max}).solve().objective})

print(pl.DataFrame(rows))

shape: (4, 2)
┌───────┬───────────┐
│ wind  ┆ objective │
│ ---   ┆ ---       │
│ f64   ┆ f64       │
╞═══════╪═══════════╡
│ 80.0  ┆ 10500.0   │
│ 120.0 ┆ 6900.0    │
│ 160.0 ┆ 5300.0    │
│ 200.0 ┆ 4500.0    │
└───────┴───────────┘

Each step of 40 saves less than the one before, because more of the wind is left over when the load is low.

diagnostics counts what the model did:

after_numbers = model.diagnostics()
print(f'{after_numbers.loads} load, {after_numbers.solves} solves')

1 load, 5 solves

The solver loaded the model once and solved it five times. The answer is the one a new solve gives:

fresh = sps.solve('dispatch.yaml', sources | {'p_max': p_max})
print(fresh.objective)

4500.0

A longer horizon¶

Now solve two days. The snapshot labels and the load grow to eight rows, and the capacity goes back to the data:

two_days = pl.DataFrame({'snapshot': range(8), 'value': 2 * load['value'].to_list()})

longer = model.update({'snapshot': two_days['snapshot'], 'load': two_days, 'p_max': sources['p_max']}).solve()
after_rows = model.diagnostics()

print(longer.objective)
print(f'{after_rows.loads} loads, {after_rows.solves} solves')

21000.0
2 loads, 6 solves

Two days cost twice one day. New snapshots add new columns, so the solver loaded the model a second time. Making a loop of solves faster starts the new solve from an earlier answer.

Next¶

In the next tutorial you solve a horizon too large for one solve, a window at a time.