This work proposes a novel method for multi-zone optimization of TCLs with latent variables. A multi-task learning-based framework is formulated to learn latent variables and models representing the time-coupled relationship. Model-based and model-free algorithms are proposed to solve the latent variable-based optimization problem.
Codes for submitted Paper "Dimension-reduced Optimization of Multi-zone Thermostatically Controlled Loads".
Authors: Xueyuan Cui, Yi Wang, and Bolun Xu.
Python version: 3.8.17
The must-have packages can be installed by running
pip install requirements.txt
All the data for experiments can be downloaded from Google Drive.
To reproduce the experiments of the proposed methods and comparisons ('OptIden', 'OptSim', 'OriIden', and 'OriSim'), please run
cd Codes/
python Lat_MB.py
python Lat_MF.py
python Ori_MB.py
python Ori_MF.py
To reproduce the experiments of generating latent and original models, please run
cd Codes/
python Lat_model.py
python Ori_model.py
To reproduce the experiments of ground-truth results, please run
cd Codes/
python Ground_truth.py
Note: There is NO multi-GPU/parallelling training in our codes.
The trained models and all figures are saved in Results. Please refer to readme.md in the Results fold for more details.