E+ and Reinforcement Learning - approach to framework advice

asked 2021-02-17 22:22:03 -0600

mechyai gravatar image

updated 2021-02-18 08:18:35 -0600

Hello everyone! For a 'little' research project, I am working on developing/analyzing reinforcement learning (RL) algorithms for HVAC and building control and energy optimization using E+ simulator, and my testbed is a vertical farming facility.

This is a new project and I am completely new to E+ and anything alike. I started reading about E+ and scoping out the available tools just a few weeks ago. My current focus is reviewing existing research/technology for RL and E+ (there are several papers). I am realizing there is a substantial learning curve for all of this.

To use E+ for an RL framework, I need to access the 'state' (read temp, humidity sensors, weather, etc.) of the running simulator and 'act' on it (actuate ventilation, fans, setpoints, etc.) at sub-hourly timesteps - highest frequency is ideal.

I am 'aware' of OpenStudio, OpenStudio Coalition, SketchUp plugin, Euclid, Modelica, BCVTB, EMS, Python EMS API, RL Testbed for E+ (IBM), and Spawn (yet to come?); but have yet to really dig into any serious documentation. I don't know where to begin.

I am looking for any advice on how to approach this project and learning in a logical manner so as to avoid reading 100s, 1000s of pages of documentation serially. I know I have lots of reading/learning ahead of me, but I can't afford to waste days/weeks digging into something that doesn't align with my project objectives.

Some foresight would be greatly appreciated. Thank you!

edit retag flag offensive close merge delete


Thierry Nouidui gravatar image Thierry Nouidui  ( 2021-02-22 00:29:25 -0600 )edit