Hi all,
I am working with a awning control scenario controlled by a reinforcement learning agent and think we are facing issues when an agent changes dinamically properties of elements of the geometry. In particular, the transmittance of some of the elements of the geometry is controlled by an actuator.
Although the IDF has initially a constant value for the schedule, the RL agent is expected to vary this schedule to simulate awning control in a house. After training and executing the agent, we find that zone temperature, humity and ratiation rate, PMV values are governed by the initial value set up in the IDF file, instead of the values controlled by the agent.
Considering the training and execution of the RL environment is not interacting with WARM-UP iterations, we suspect geometry-based calculations become STATIC and are not affected by the agent - but any hint on this would be welcome as we are not sure about it.
I also tried involving the agent during warm-up-days phases but I noticed the following:
- PMV values are not available during design days.
- Solar radiation rates are constant during executions regardless of agent actions on awning actuators - that is, calculations may not be considering the effects of awnings on the zone solar radiation rate.
Do you think the intended scenario beyond current E+ simulation capabilities or do you think we taking an incorrect approach to the problem?
Any feedback or comment would be highly welcome!
Thanks in advance
