Autarky Mini-Grid Optimization Models

Overview

Autarky is a modern open-source framework for the optimal sizing and dispatch of decentralized mini-grid systems, tailored for rural electrification and energy resilience. Implemented in Julia with JuMP, it supports hybrid systems including solar PV, batteries, diesel generators, and grid connection under both deterministic and stochastic conditions.

The framework enables robust and cost-effective energy system design by minimizing the Net Present Cost (NPC), accounting for capital investment, replacement, operational costs, and salvage value all under a multi-year horizon with optional seasonality.

Autarky supports four distinct optimization formulations:

• Regular (Deterministic): Least-cost sizing and dispatch assuming perfect foresight.
• Expected Value Model (EVM): Incorporates uncertainty using expected forecasting errors.
• Individual Chance Constraints (ICC): Enforces per-time-step reliability under uncertainty.
• Joint Chance Constraints (JCC): Ensures system-wide reliability across outage windows, the most robust approach.

Key Features

• Hybrid energy system modeling: PV, wind, battery, diesel, grid.
• Modular activation of components and constraints.
• Seasonal time series support.
• Reserve planning under outages and forecast uncertainty.
• Advanced reliability modeling with probabilistic constraints.
• Objective: Minimize NPC via discounted cash flow logic.
• Optional unit commitment formulation with integer sizing.

Repository Structure

Each model variant (Regular, EVM, ICC, JCC) resides in a dedicated folder with the following structure:

```bash

autarky/

deterministic/ # Deterministic model expectedvalues/ # EVM with forecast errors icc/ # Individual Chance Constraints jccgenz/ # Joint Chance Constraints

Each model folder contains: src/ main.jl # Entrypoint to run the optimization inputs/ *.csv, *.yaml # Time-series and techno-economic parameters results/ *.csv # Output results saved here

```

To run a model:

bash cd <selected model>/src julia main.jl

Inputs

• inputs/parameters.yaml: General project and technology configuration
• CSV time-series:
  • load.csv: Load demand profile
  • solar_unit.csv: Unit production of PV
  • wind_unit.csv: (Optional) Wind production
  • gridcost.csv, gridprice.csv: Grid tariffs (optional)
  • grid_availability.csv: Binary grid outage series (for stochastic modeling)
  • solarerrors.csv, loaderrors.csv: Forecasting error samples (for EVM/ICC/JCC)

Model Comparison

| Model | Description | Reliability Scope | Complexity | Runtime | Robustness | |------------------|------------------------------------------------------------------------------------|--------------------------------------------------|-------------------------------|-----------------------|-------------------| | Regular | Least-cost, deterministic optimization assuming perfect foresight | None (0%) | Linear | Fast | Low | | EVM | Penalizes expected shortfall using nonlinear cost terms | Expected-value reliability (~20%) | Nonlinear | Moderate | Medium | | ICC | Enforces a confidence level per time step using normal quantiles | Per-timestep reliability (~60%) | Nonlinear | Moderate | Medium | | JCC | Guarantees feasibility across outage windows using joint probability constraints | Window-based reliability (90% over hours) | Nonlinear + Multivariate | Minutes | High |

Streamlit Viewer App

Autarky comes with a Streamlit web app to visualize and compare projects:

bash cd autarky/app streamlit run app.py

Features:

• Visualize Inputs: Time series, component specs, costs, error metrics
• Visualize Results: Dispatch, cost breakdown, sizing, LCOE, NPC
• Compare Projects: Assess sizing and cost trade-offs across models

Workflow:

• Run a model (main.jl) in one of the four folders.
• Copy the generated inputs/ and results/ folders.
• Paste them into a new folder inside Autarky App/projects/yourprojectname/.

Now you can explore the project via the Streamlit UI.

Requirements

• Julia 1.9
• Packages: JuMP, Ipopt, GLPK, Distributions, YAML, etc.
• Python 3.10 for the Streamlit app
• Streamlit dependencies: pandas, plotly, matplotlib, etc.

Citation

This framework builds on the methods described in: - Ouanes, N., Gonzlez Grandn, T., Heitsch, H., Henrion, R. (2025). Optimizing the economic dispatch of weakly-connected mini-grids under uncertainty using joint chance constraints. Annals of Operations Research.