For more details, please see our paper Predicting Strategic Energy Storage Behaviors which has been accepted at IEEE transaction on Smart Grid, 2023. If this is useful for your work, please cite our paper.
This repository contains the code to reproduce the results.
We propose a gradient-based approach to identify the strategic energy storage behavior model, where the agent is to minimize the energy cost and a disutility cost and satisfying the energy constraints and State-of-charge constraints. We assume the agent disutility function form is not known, thus we propose to use input convex neural networks (ICNN) to capture the disutility function.
cd quadraticenergystorage
# Running the baselines (MLP or RNN)
python baseline.py --model MLP
# Running generic model
python train_general.py
We also compare the proposed approach using optimization-tools to directly solving the single-level optimization, please see optimization.m for more detail.
cd genericenergystorage
# Running the baselines (MLP or RNN)
python baseline.py --model MLP
# Running generic model
python train_general.py
# Running quadratic model
python train_quadratic.py
cd Tesla
# Running the baselines (MLP or RNN)
python baseline.py --model MLP
# Running generic model
python train_general.py
# Running quadratic model
python train_quadratic.py