Skip to content


Switch branches/tags

Name already in use

A tag already exists with the provided branch name. Many Git commands accept both tag and branch names, so creating this branch may cause unexpected behavior. Are you sure you want to create this branch?

Latest commit


Git stats


Failed to load latest commit information.
Latest commit message
Commit time

unit tests

Project Description

Reinforcement Learning Testbed for Power Consumption Optimization.

Contributing to the project

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 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.

Supported platforms

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.

Building from source

Installation of rl-testbed-for-energyplus consists of three parts:

  • Install EnergyPlus prebuilt package
  • Build patched EnergyPlus
  • Install built executables

Install EnergyPlus prebuilt package

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:

Linux MacOS
8.8.0 EnergyPlus-8.8.0-7c3bbe4830-Darwin-x86_64.dmg
9.1.0 EnergyPlus-9.1.0-08d2e308bb-Darwin-x86_64.dmg
9.2.0 EnergyPlus-9.2.0-921312fa1d-Darwin-x86_64.dmg
9.3.0 EnergyPlus-9.3.0-baff08990c-Darwin-x86_64.dmg
9.4.0 EnergyPlus-9.4.0-998c4b761e-Darwin-macOS10.15-x86_64.dmg
9.5.0 EnergyPlus-9.5.0-de239b2e5f-Darwin-macOS11.2-arm64.dmg

You can also download the installer at

  1. Go to the web page shown above.
  2. Right click on relevant link in supported versions table and select Save link As to from the menu to download installation image.
  3. (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)
$ patch -p0 < rl-testbed-for-energyplus/EnergyPlus/
  1. Execute installation image. Below example is for EnergyPlus 9.1.0

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>.

  1. Go to the web page shown above.
  2. Right click in supported versions table and select Save link As to from the menu to download installation image.
  3. Double click the downloaded package, and follow the instructions. The package will be installed in /Applications/EnergyPlus-<EPLUS_VERSION>.

Build patched EnergyPlus

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)

$ git clone -b v<EPLUS_VERSION>
$ git clone

Apply patch to EnergyPlus and build. Replace <EPLUS_VERSION> by the one you're using (for instance, 9-3-0)

$ 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

Install built executables

$ sudo make install

Install Python dependencies

Python3 >= 3.8 is required.

OpenAI Baselines
$ 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 for details.

Older versions:

To run on Ubuntu 18.04, you'll need the following pip dependencies:

Ray RLlib
$ pip3 install -r requirements/ray.txt

How to run

Set up

Some environment variables must be defined. ENERGYPLUS_VERSION must be adapted to your version.

In $(HOME)/.bashrc

# Specify the top directory

if [ `uname` == "Darwin" ]; then
export ENERGYPLUS="${ENERGYPLUS_DIR}/energyplus"

# 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_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.

# 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

OpenAI Baselines

$ time python3 -m baselines_energyplus.trpo_mpi.run_energyplus --num-timesteps 1000000000

Ray RLlib

$ 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
- -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.

EnergyPlus monitor

Several graphs are shown.

  1. Zone temperature and outdoor temperature
  2. West zone return air temperature and west zone setpoint temperature
  3. Mixed air, fan, and DEC outlet temperatures
  4. IEC, CW, DEC outlet temperatures
  5. Electric demand power (whole building, facility, HVAC)
  6. 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.

How to cite

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:

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",
publisher="Springer Singapore",

Related information


Reinforcement Learning Testbed for Power Consumption Optimization using EnergyPlus







No releases published


No packages published