For more details, please see our paper Demand Response Model Identification and Behavior Forecast with OptNet: a Gradient-based Approach which has been accepted at ACM E-energy, 2022. If this is useful for your work, please cite our paper:

@inproceedings{bian2022demand,
  title={Demand response model identification and behavior forecast with OptNet: a gradient-based approach},
  author={Bian, Yuexin and Zheng, Ningkun and Zheng, Yang and Xu, Bolun and Shi, Yuanyuan},
  booktitle={Proceedings of the Thirteenth ACM International Conference on Future Energy Systems},
  pages={418--429},
  year={2022}
}

Install the package

pip install -r requirements.txt

• ESID-data: NYISO 2019 real-time electricity price data.
• Results: Contain 10 random generated data: data1-data10. Figures and post-processing data also in this folder.
• data_generation.py: generate random ground truth parameters. Random select dates in real-time price data to generate true dispatch data for training and validation.
• main.py: using OptNet to learn parameters with training data, check validation loss with learned parameters.
• MLP.py: baseline, using three-layer forward ReLU network, return validation losses of differet data and iteration numbers.
• plot.py: make figures for energy storage models.
• post_processing.py: using differernt iteration numbers OptNet learned parameters to calculate validation loss.
• utils.py: functions used in other scripts.

• dataset: contain price and ambient data, from Fernández-Blanco, R., Morales, J. M., & Pineda, S. (2020). Forecasting the Price-Response of a Pool of Buildings via Homothetic Inverse Optimization (https://github.com/groupoasys/homothetic)
• baseline_NN.py: Neural network as baseline to predict the power consumation
• gradient_method.py: Our approach
• utils.py: contain the solve function and etc.

You can use --num to determine the number of experiments, --choice to determine whether the model learns the indoor temperature, --save to determine whether to save the identified model parameters.

python gradient_method.py # default experiment setting
python gradient_method.py --num 10 --save True --seed 1234 --choice 2

• dataset: contain all the input data in the experiments
• baseline_IO.py: reproducing code for Inverse optimization approach to the identification of electricity consumer models
• baselines.py: RNN and MLP model
• gradient_method.py: our model for DR identification

You can use --K to determine the number of loads, --noise to determine the noise level, --save to determine whether to save the identified model parameters.

python gradient_method.py # default experiment setting
python gradient_method.py --K 3 --noise 2 --save True

For baseline comparison, you can use the following method to save the identified model parameters.

python baseline_IO.py --K --noise 2 --save True