Is it a good idea to stop training if there is a violation of a hard constraint in reinforcement learning?

2 ビュー (過去 30 日間)
Aysegul Kahraman
Aysegul Kahraman 2022 年 3 月 17 日
I have a physical model in Simulink and I am trying to do the scheduling for my components by using RL toolbox.
There are some constraints like the one in 'Water Distribution System Scheduling' example. For example, for the tank example, the water level can not go lower than 0 or beyond the level of the tank. This example also uses the stopping simulation approach. However, if I train my model in this way, during the simulation it might still go beyond the limits and stop the simulation before the actual simulation time ends, which means the schedule for the dayahead is not completed.
What is the best method for maintaining hard constraints or physical boundaries with RL?
Any comments would be appreciated.

回答 (0 件)

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by