Skip to content

Time Reduction Settings

PowerGenome can reduce a full 8760-hour year to a smaller set of representative periods — groups of consecutive days that stand in for the full year. This reduces the size of your optimization problem while preserving the statistical properties of demand and variable generation.

Time domain reduction is applied jointly to load profiles and variable generation profiles (wind, solar) using k-means clustering.

Parameters

reduce_time_domain

Type: bool Default: false

Set to true to enable time domain reduction. When false (or absent), all 8760 hours are passed to GenX as a single period with weight 8760.

Time reduction only works on data with ≤ 8760 hourly rows. If your data is already shorter (e.g., a previous reduction), the step is skipped automatically.

reduce_time_domain: true

time_domain_periods

Type: int Required when: reduce_time_domain: true

The number of representative periods (clusters) to select. Each period is a block of time_domain_days_per_period consecutive days.

Common values:

Value Description
4 Seasonal quarters (very coarse)
12 Monthly-style reduction
52 Weekly — one week per calendar week 
time_domain_periods: 12

time_domain_days_per_period

Type: int Required when: reduce_time_domain: true

The number of consecutive days in each representative period. Combined with time_domain_periods, this defines the total number of hours in your reduced time series:

$$\text{total hours} = \text{time_domain_periods} \times \text{time_domain_days_per_period} \times 24$$

For a model with unit-commitment constraints, using multiple days per period (e.g., 7) is recommended so that start-up and shut-down decisions span realistic multi-day sequences.

time_domain_days_per_period: 7   # weekly periods

include_peak_day

Type: bool Required when: reduce_time_domain: true

When true, the day with the highest system-wide demand is always included as one of the representative periods, regardless of which day k-means would select. This ensures the optimization model sees the true system peak.

include_peak_day: true

demand_weight_factor

Type: int or float Required when: reduce_time_domain: true

Before k-means clustering, all profiles (load and variable generation) are scaled to the range [0, 1]. Demand profiles are then multiplied by this factor. A value greater than 1 weights demand more heavily than renewable generation when selecting clusters.

  • 1: demand and variable generation weighted equally
  • 5: demand contributes 5× more to cluster selection (recommended starting point)
demand_weight_factor: 5

tdr_n_init

Type: int Default: 100 Optional

Number of k-means initializations to run. More initializations reduce the chance of converging to a local minimum, at the cost of longer run time.

tdr_n_init: 100

Minimal example

reduce_time_domain: true
time_domain_periods: 12
time_domain_days_per_period: 7
include_peak_day: true
demand_weight_factor: 5

This creates 12 representative weeks, with the peak demand day guaranteed to appear in one of them.

Output format

When time reduction is active, the system/Demand_data.csv output (or Load_data.csv when using the legacy old_genx_format) includes additional metadata columns that GenX uses to reconstruct the weight of each representative period:

Column Description
Rep_Periods Total number of representative periods
Timesteps_per_Rep_Period Hours per period (days × 24)
Sub_Weights Number of hours from the full year that each period represents
Time_Index Hour index (1 to Rep_Periods × Timesteps_per_Rep_Period)

Sub_Weights must sum to 8760 across all periods for an annual model.

Choosing parameter values

There is no universally correct number of representative periods. Typical guidance:

  • 12–52 periods is a common range for capacity expansion. Fewer periods run faster but miss nuance in correlated wind/solar/demand patterns.
  • 7-day periods are a good starting point. Shorter periods (1–2 days) can cause the optimizer to ignore multi-day operational constraints.
  • Always include the peak day when your model has capacity reserve constraints.
  • Check convergence: run with a larger number of periods and verify that model results are stable before reducing further.