The Value of Storage in Electricity Distribution: The Role of Markets

This repository supplements the paper "The value of storage in electricity distribution: The role of markets" by Dirk Lauinger, Deepjyoti Deka, and Sungho Shin, available at https://arxiv.org/abs/2510.12435, and funded by the Future Energy Systems Center at the MIT Energy Initiative.

Repository Structure

The repository is organized as follows:

• code/ Data structure, optimization model, and experiment scripts;
• data/ Input data and preprocessing scripts;
• pics/ Figures reproduced in the paper;
• results/ Output files from numerical experiments.

Reproducing the Numerical Experiments

Software Requirements

The numerical experiments were implemented using:

• Julia 1.11.2
• JuMP 1.23.5
• Gurobi 12.0.2

A valid Gurobi license is required to run the optimization model. To install Julia dependencies, run this command in bash

sh start_julia.sh

followed by this command in Julia

]
instantiate

Hardware Setup

All numerical experiments were conducted on AMD EPYC 9474F CPUs with 48 cores, a 3.6GHz base clock, and 376GB of RAM. On this hardware, running all nine experiments in Table C1 sequentially took 18 hours.

Running the Experiments

Run the experiments as follows:

sh code/ex1.sh
...
sh code/ex9.sh

Results are written to results/experiments/.

Analyzing the Results

Obtain the data in Table C1 for each experiment by running:

sh code/06_run_analysis_experiment.sh ex1
...
sh code/06_run_analysis_experiment.sh ex9

Code Description

The folder code/ contains:

• 01_data.jl: Generates and populates a data structure for the investment planning optimization problem (B1);
• 02_peak_shaving_potential.jl: Generates the data underlying Figure A3 and can be used to determine the upper bound $\bar x_\mathrm{s}$ on storage investment levels;
• 03_model.jl: Numerical implementation of problem (B1);
• 04_analysis.jl: Analyzes the results of the numerical experiments and compiles Table C1, generates the data underlying Figures 5, 6, and C1, and generates Figure C1;
• 05_run_plan.jl: Runs an instance of the planning problem specified by command line arguments;
• 05_run_plan.sh: Runs 05_run_plan.jl;
• 06_run_analysis_experiment.jl: Runs 04_analysis.jl with a command line argument;
• 06_run_analysis_experiment.sh: Runs 06_run_analysis_experiment.jl;
• ex1.sh ... ex9.sh: Run the numerical experiments in Table C1.

Data Description

The folder data/ contains all datasets used in the numerical experiments.

Nantucket Electricity Demand and Prices

ISONE_data/ contains

nodalloadweights_4006_202401.csv
...
nodalloadweights_4006_202412.csv

These files are available from the ISO New England.

The data is analyzed in data_analysis_2024.ipynb and used to create Figure 2 and the data file Nantucket_2024.csv, which is used by code/01_data.jl when populating the data structure for the investment planning problem.

US Installed Generation Capacity

eia-860/ contains

december_generator2025.xlsx

This file is available from the Preliminary Monthly Electric Generator Inventory by the US Energy Information Administration and analyzed in storage_eia860.ipynb to compute the numbers on installed US battery storage in Section A.1 of the paper.

Battery Cost Projections

nrel_battery_cost/ contains

2019.txt
...
2025.txt

These files represent NREL cost projections on utility-scale battery storage. The 2025 projections are available at https://docs.nrel.gov/docs/fy25osti/93281.pdf. The data is analyzed in nrel_battery_costs.ipynb to create Figure A1.

Other Data

data_to_json.jl builds the data file nantucket.json for the numerical case study. This file is read by code/01_data.jl when populating the data structure for the investment planning problem.

Figures

The folder pics/ contains figures reproduced in the paper:

• battery_cost.pdf and battery_cost.svg for Figure A1;
• heatmap.pdf for Figure 6;
• load_24.pdf, load_24.svg, price_24.pdf, and price_24.svg for Figure 2.

Results

The folder results/ contains outputs from numerical experiments:

• experiments/ contains .jld and .log files for each of the nine numerical experiments;
• planning/ is empty and used by default to log results that do not correspond to any of the nine experiments;
• potential.txt shows the maximum peak shaving potential and the battery power and energy capacity required to achieve that potential for various roundtrip efficiencies in the Nantucket case study. This data is used to generate Figure A3.

Version

Release v0.1.0-submission is the version of the repository used in the inital submission of the paper for peer review.