Stochastic Programming#

The Stochastics extension in Temoa v4 provides support for stochastic programming using the mpi-sppy library. This allows for decision-making under uncertainty by considering multiple scenarios simultaneously and finding an optimal “first-stage” decision that minimizes the expected cost over all scenarios.

Stochastic programming is particularly useful for modeling uncertainties in future costs, demands, or resource availability.

Dependencies#

The stochastics extension requires the mpi-sppy package. You can install it using uv:

uv add mpi-sppy

Or using pip:

pip install mpi-sppy

Configuration#

To run Temoa in stochastic mode, you need to modify your main configuration TOML file and provide an additional stochastic configuration file.

Main Configuration TOML#

Set the scenario_mode to "stochastic" and add a [stochastic] section:

scenario_mode = "stochastic"

# ... other standard options ...

[stochastic]
# Path to the stochastic configuration file, relative to this file
stochastic_config = "stochastic_config.toml"

Stochastic Configuration TOML#

The stochastic configuration file defines the scenarios, their probabilities, and the data perturbations associated with each scenario.

# Define the scenarios
[scenarios]
# Each key is a scenario name, and the value is its probability
# Probabilities must sum to 1.0
low_cost = 0.5
high_cost = 0.5

# Define perturbations for a specific scenario
[[perturbations]]
scenario = "low_cost"
table = "cost_variable"
# Filter specifies which rows in the table to perturb
filter = { tech = "IMPHCO1" }
# Action can be "multiply", "add", or "set" (defaults to "set")
action = "multiply"
value = 0.5

[[perturbations]]
scenario = "high_cost"
table = "cost_variable"
filter = { tech = "IMPHCO1" }
action = "multiply"
value = 1.5

Perturbation Options#

Currently, the following fields are required for each perturbation:

scenario: The name of the scenario to which this perturbation applies.
table: The Temoa parameter (database table) to perturb (e.g., cost_variable, demand, capacity_factor_process).
filter: A dictionary of column-value pairs used to identify specific rows. Since the extension uses the dynamic manifest from HybridLoader, any column belonging to the table’s index can be used for filtering.
action: The operation to perform. Supported values:
- multiply: Multiply the base value by value.
- add: Add value to the base value.
- set: Replace the base value with value.
value: The numeric value used in the perturbation action.

How it Works#

When running in stochastic mode, Temoa:

Loads the base data from the input database.
Identifies the “first-stage” variables. In the current implementation, all decisions in the first time period are considered first-stage.
Orchestrates multiple scenario runs using the mpi-sppy Extensive Form (EF) solver.
For each scenario, the scenario_creator applies the specified perturbations to the base data and builds a Pyomo model instance.
The EF solver binds the first-stage variables across all scenarios (non-anticipativity constraints) and optimizes the total expected cost.
The terminal output reports the Stochastic Expected Value.

Limitations#

Two-Stage Only: While mpi-sppy supports multi-stage stochastic programming, the current Temoa integration is tailored for two-stage problems where the first time period constitutes the first stage.
Result Persistence: Currently, only the expected objective value and summary logs are produced. Detailed per-scenario result persistence to the database is under development.

Stochastic Programming

Contents