Reinforcement Learning Testbed for Power Consumption Optimization.
We welcome contributions to this project in many forms. There's always plenty to do! Full details of how to contribute to this project are documented in the CONTRIBUTING.md file.
The project's maintainers: are responsible for reviewing and merging all pull requests and they guide the over-all technical direction of the project.
We have tested on the following platforms.
- macOS High Sierra (Version 10.13.6)
- macOS Catalina (Version 10.15.3)
- Ubuntu 20.04 LTS
The easiest way to setup a training environment is to use the docker image. See instructions here. For manual installation, see below.
Installation of rl-testbed-for-energyplus consists of three parts:
- Install EnergyPlus prebuilt package
- Build patched EnergyPlus
- Install built executables
First, download pre-built package of EnergyPlus and install it. This is not for executing normal version of EnergyPlus, but to get some pre-compiled binaries and data files that can not be generated from source code.
Supported EnergyPlus versions:
and up to EnergyPlus 22.2.0. See EnergyPlus folder which provides a list of available patches.
You can also download the installer at https://github.com/NREL/EnergyPlus/releases/.
- Go to the web page shown above.
- Right click on relevant link in supported versions table and select
Save link Asto from the menu to download installation image. - (9.1.0, Linux only) Apply patch on downloaded file (EnergyPlus 9.1.0 installation script unpacks in /usr/local instead of /usr/local/EnergyPlus-9.1.0)
$ cd <DOWNLOAD-DIRECTORY>
$ patch -p0 < rl-testbed-for-energyplus/EnergyPlus/EnergyPlus-9.1.0-08d2e308bb-Linux-x86_64.sh.patch
- Execute installation image. Below example is for EnergyPlus 9.1.0
$ sudo bash <DOWNLOAD-DIRECTORY>/EnergyPlus-9.1.0-08d2e308bb-Linux-x86_64.sh
Enter your admin password if required.
Specify /usr/local for install directory.
Respond with /usr/local/bin if asked for symbolic link location.
The package will be installed at /usr/local/EnergyPlus-<EPLUS_VERSION>.
- Go to the web page shown above.
- Right click in supported versions table and select
Save link Asto from the menu to download installation image. - Double click the downloaded package, and follow the instructions.
The package will be installed in
/Applications/EnergyPlus-<EPLUS_VERSION>.
Download source code of EnergyPlus and rl-testbed-for-energyplus. In below scripted lines, replace <EPLUS_VERSION>
by the one you're using (for instance, 9.3.0)
$ cd <WORKING-DIRECTORY>
$ git clone -b v<EPLUS_VERSION> [email protected]:NREL/EnergyPlus.git
$ git clone [email protected]:ibm/rl-testbed-for-energyplus.git
Apply patch to EnergyPlus and build. Replace <EPLUS_VERSION>
by the one you're using (for instance, 9-3-0)
$ cd <WORKING-DIRECTORY>/EnergyPlus
$ patch -p1 < ../rl-testbed-for-energyplus/EnergyPlus/RL-patch-for-EnergyPlus-<EPLUS_VERSION>.patch
$ mkdir build
$ cd build
$ cmake -DCMAKE_INSTALL_PREFIX=/usr/local/EnergyPlus-<EPLUS_VERSION> .. # Ubuntu case (please don't forget the two dots at the end)
$ cmake -DCMAKE_INSTALL_PREFIX=/Applications/EnergyPlus-<EPLUS_VERSION> .. # macOS case (please don't forget the two dots at the end)
$ make -j4
$ sudo make install
Python3 >= 3.8 is required.
$ pip3 install -r requirements/baselines.txt
Main dependencies:
- tensorflow 2.5
- baselines 0.1.6
- gym 0.15.7
Note on baselines dependency:
- baselines 0.1.5 fails to install when MuJoCo can't be found. Reason why 0.1.6 is required (available from sources only)
- if baselines 0.1.6 installation fails because TensorFlow is missing, install tensorflow manually first, then retry.
For more information on baselines requirements, see https://github.com/openai/baselines for details.
Older versions:
To run on Ubuntu 18.04, you'll need the following pip dependencies:
scipy==1.5.4
tensorflow==1.15.4
$ pip3 install -r requirements/ray.txt
Some environment variables must be defined. ENERGYPLUS_VERSION must be adapted to your version.
In $(HOME)/.bashrc
# Specify the top directory
TOP=<DOWNLOAD-DIRECTORY>/rl-testbed-for-energyplus
export PYTHONPATH=${PYTHONPATH}:${TOP}
if [ `uname` == "Darwin" ]; then
energyplus_instdir="/Applications"
else
energyplus_instdir="/usr/local"
fi
ENERGYPLUS_VERSION="8-8-0"
#ENERGYPLUS_VERSION="9-1-0"
#ENERGYPLUS_VERSION="9-2-0"
#ENERGYPLUS_VERSION="9-3-0"
ENERGYPLUS_DIR="${energyplus_instdir}/EnergyPlus-${ENERGYPLUS_VERSION}"
WEATHER_DIR="${ENERGYPLUS_DIR}/WeatherData"
export ENERGYPLUS="${ENERGYPLUS_DIR}/energyplus"
MODEL_DIR="${TOP}/EnergyPlus/Model-${ENERGYPLUS_VERSION}"
# Weather file.
# Single weather file or multiple weather files separated by comma character.
export ENERGYPLUS_WEATHER="${WEATHER_DIR}/USA_CA_San.Francisco.Intl.AP.724940_TMY3.epw"
#export ENERGYPLUS_WEATHER="${WEATHER_DIR}/USA_CO_Golden-NREL.724666_TMY3.epw"
#export ENERGYPLUS_WEATHER="${WEATHER_DIR}/USA_FL_Tampa.Intl.AP.722110_TMY3.epw"
#export ENERGYPLUS_WEATHER="${WEATHER_DIR}/USA_IL_Chicago-OHare.Intl.AP.725300_TMY3.epw"
#export ENERGYPLUS_WEATHER="${WEATHER_DIR}/USA_VA_Sterling-Washington.Dulles.Intl.AP.724030_TMY3.epw"
#export ENERGYPLUS_WEATHER="${WEATHER_DIR}/USA_CA_San.Francisco.Intl.AP.724940_TMY3.epw,${WEATHER_DIR}/USA_CO_Golden-NREL.724666_TMY3.epw,${WEATHER_DIR}/USA_FL_Tampa.Intl.AP.722110_TMY3.epw"
# Ouput directory "openai-YYYY-MM-DD-HH-MM-SS-mmmmmm" is created in
# the directory specified by ENERGYPLUS_LOGBASE or in the current directory if not specified.
export ENERGYPLUS_LOGBASE="${HOME}/eplog"
# Model file. Uncomment one.
#export ENERGYPLUS_MODEL="${MODEL_DIR}/2ZoneDataCenterHVAC_wEconomizer_Temp.idf" # Temp. setpoint control
export ENERGYPLUS_MODEL="${MODEL_DIR}/2ZoneDataCenterHVAC_wEconomizer_Temp_Fan.idf" # Temp. setpoint and fan control
# Run command (example)
# $ time python3 -m baselines_energyplus.trpo_mpi.run_energyplus --num-timesteps 1000000000
# Monitoring (example)
# $ python3 -m common.plot_energyplus
Simulation process starts by the following command. The only applicable option is --num-timesteps
$ time python3 -m baselines_energyplus.trpo_mpi.run_energyplus --num-timesteps 1000000000
$ time python3 -m ray_energyplus.ppo.run_energyplus --num-timesteps 1000000000
Output files are generated under the directory ${ENERGYPLUS_LOGBASE}/openai-YYYY-MM-DD-HH-MM-SS-mmmmmm. These include:
- log.txt Log file generated by baselines Logger.
- progress.csv Log file generated by baselines Logger.
- output/episode-NNNNNNNN/ Episode data
Epsiode data contains the following files:
- 2ZoneDataCenterHVAC_wEconomizer_Temp_Fan.idf A copy of model file used in the simulation of the episode
- USA_CA_San.Francisco.Intl.AP.724940_TMY3.epw A copy of weather file used in the simulation of the episode
- eplusout.csv.gz Simulation result in CSV format
- eplusout.err Error message. You need make sure that there are no Severe errors
- eplusout.htm Human readable report file
You can monitor the progress of the simulation using plot_energyplus utility.
$ python3 -m common.plot_energyplus
Options:
- -l <log_dir> Specify log directory (usually openai-YYYY-MM-DD-HH-MM-SS-mmmmmm)
- -c <csv_file> Specify single CSV file to view
- -d Dump every timestep in CSV file (dump_timesteps.csv)
- -D Dump episodes in CSV file (dump_episodes.dat)
If neither -l nor -c option is specified, plot_energyplus tries to open the latest directory under ${ENERGYPLUS_LOG} directory.
If none of -d or -D is specified, the progress windows is opened.
Several graphs are shown.
- Zone temperature and outdoor temperature
- West zone return air temperature and west zone setpoint temperature
- Mixed air, fan, and DEC outlet temperatures
- IEC, CW, DEC outlet temperatures
- Electric demand power (whole building, facility, HVAC)
- Reward
Only the current episode is shown in the graph 1 to 5. The current episode is specified by pushing one of "First", "Prev", "Next", or "Last" button, or directly clicking the appropriate point on the episode bar at the bottom. If you're at the last episode, the current episode moves automatically to the latest one as new episode is completed.
Note: The reward value shown in the graph 6 is retrieved from "progress.csv" file generated by TRPO baseline, which is not necessarily same as the reward value computed by our reward function.
You can pan or zoom each graph by entering pan/zoom mode by clicking cross-arrows on the bottom left of the window.
When new episode is shown on the window, some statistical information is show as follow:
episode 362
read_episode: file=/home/moriyama/eplog/openai-2018-07-04-10-48-46-712881/output/episode-00000362/eplusout.csv.gz
Reward ave= 0.77, min= 0.40, max= 1.33, std= 0.22
westzone_temp ave=22.93, min=21.96, max=23.37, std= 0.19
eastzone_temp ave=22.94, min=22.10, max=23.51, std= 0.17
Power consumption ave=102,243.47, min=65,428.31, max=135,956.47, std=18,264.50
pue ave= 1.27, min= 1.02, max= 1.63, std= 0.13
westzone_temp distribution
degree 0.0-0.9 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
-------------------------------------------------------------------------
18.0C 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0%
19.0C 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0%
20.0C 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0%
21.0C 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0%
22.0C 50.8% 0.0% 0.1% 0.7% 3.4% 5.6% 2.2% 0.7% 1.0% 0.9% 36.4%
23.0C 49.2% 49.0% 0.2% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0%
24.0C 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0%
25.0C 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0%
26.0C 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0%
27.0C 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0%
The reward value shown above is computed by applying reward function to the simulation result.
The Reinforcement Learning Testbed for Power Consumption Optimization Project uses the MIT License software license.
For citing the use or extension of this testbed, you may cite our paper at AsiaSim 2018, which can be found at Springer or as a slightly revised version at Arxiv. You may use the following BibTeX entry:
@InProceedings{10.1007/978-981-13-2853-4_4,
author="Moriyama, Takao and De Magistris, Giovanni and Tatsubori, Michiaki and Pham, Tu-Hoa and Munawar, Asim and Tachibana, Ryuki",
title="Reinforcement Learning Testbed for Power-Consumption Optimization",
booktitle="Methods and Applications for Modeling and Simulation of Complex Systems",
year="2018",
publisher="Springer Singapore",
address="Singapore",
pages="45--59",
isbn="978-981-13-2853-4"
}
- A pre-print version of AsiaSim2018 paper on arXiv: https://arxiv.org/abs/1808.10427
- EnergyPlus: https://github.com/NREL/EnergyPlus
- OpenAI Gym: https://github.com/OpenAI/gym
- OpenAI Baselines: https://github.com/OpenAI/baselines
- Ray: https://github.com/ray-project/ray