Heated Pipe System with Heat Loss
Interactive version of system_tutorial.ipynb: regulating the outlet temperature of a 450 m heat-pipe loop near its boiling point with PI, MPC, and Reinforcement Learning control. See the full project on GitHub.
Constant heater power, no control — shows the raw system dynamics.
SISO PI controller with gain scheduling on inlet temperature (keeps inlet below the 95°C safety limit).
Linear MPC controlling inlet & outlet temperature jointly under power and safety constraints. Each run solves ~360 QPs, so it takes noticeably longer than PI — up to a few minutes, or instant if these exact settings were already run (cached).
Pretrained Soft Actor-Critic (SAC) policy (300k training steps). alpha blends the raw action with the previous one (1 = no smoothing). Tracks the setpoint to within ~2°C overshoot, safely under the 95°C inlet limit.
Run PI, MPC, and RL under the same conditions and overlay them on one graph — live equivalent of the notebook's compare_everything.py (which plotted pre-saved CSVs). PI and MPC use their hand-tuned defaults from the notebook; only the shared conditions below are adjustable. This is the slowest simulation here (MPC dominates) — expect ~2-3 minutes, or instant if these exact conditions were already run (results are cached).