Reactor¶
Nuclear reactor — point-kinetics with delayed neutrons, xenon poisoning, thermal feedback, and rate-limited control rods.
| Action space | Box((1,)), all actions in [-1, 1] |
| Observation space | Box((4,)) |
| Tracked variable(s) | neutron power (normalised) |
| Episode length | 864 steps (8640 s at 10 s per step) |
| Import | from target_gym import Reactor, ReactorParams |
| Cite as | reactor-v2 |
Action space¶
Actions are normalised to [-1, 1] and mapped onto the plant's real
actuator range inside the environment.
| # | meaning | min | max |
|---|---|---|---|
| 0 | rho_ext_norm | -1 | 1 |
Observation space¶
4 values. Indices (0,) carry the tracked variable(s) that the reward scores.
Rewards¶
Flux tracking minus a small rod-motion penalty.
Every environment in this suite scores on one contract: the reward is
(tracking terms, multiplied) x (1 - weighted costs), bounded in
[0, 1], and reaches 1 only while the target is held exactly. See
Reward shaping.
Starting state¶
reset samples the initial condition and the target; state has 12 fields.
Episode end¶
Termination. See check_is_terminal.
Truncation. After 864 steps.
Baselines¶
A tuned PID ships with this environment.
Arguments¶
| parameter | default |
|---|---|
delta_t |
1 |
max_steps_in_episode |
864 |
Lambda_gen |
0.0001 |
alpha_fuel |
-3e-05 |
alpha_coolant |
-5e-05 |
T_fuel_ref |
900 |
T_coolant_ref |
580 |
rho_ext_min |
-0.01 |
rho_ext_max |
0.005 |
rod_speed_insert |
0.0004 |
rod_speed_withdraw |
0.0002 |
P_thermal_ref |
3e+09 |
C_fuel |
3.3e+07 |
C_coolant |
7e+07 |
| … | 22 more, see the params dataclass |