Boiler Drum

Hold the drum level and the steam pressure of a 160 MW natural-circulation drum boiler with fuel and feedwater, while the turbine's steam draw moves. Drum level is the classic non-minimum-phase loop: a pressure drop swells the level while water is leaving, and cold feedwater shrinks it before it rises. The level integrates, the riser voidage behind both effects is hidden, and both a high and a low level trip the boiler.

import target_gym as tg

env, params = tg.make("boiler_drum")

Action space	Box((2,)), all actions in [-1, 1]
Observation space	Box((7,))
Tracked variable(s)	drum level (m), drum pressure (bar)
Episode length	400 steps of 2 s (13.3 min)
Cite as	boiler_drum-v2
Physics	contract

Action space

Actions are normalised to [-1, 1] and mapped onto the plant's real actuator range inside the environment.

#	meaning	min	max
0	fuel	-1	1
1	feedwater	-1	1

Observation space

7 values in the plant's own units. The space declares no bounds (it spans the float32 range), so scale them before training. The reward scores obs[0] against obs[5], obs[1] against obs[6].

#	meaning
0	drum level relative to normal water level (m)
1	drum pressure (bar)
2	q_steam, steam flow to the turbine (kg/s), the disturbance
3	fuel, percent of maximum
4	feedwater, percent of maximum
5	target level, always 0 (m)
6	target pressure (bar)

Rewards

See the environment's compute_reward.

Every environment in this suite scores on one contract. The reward is minus the sum of three non-negative costs, -(tracking_cost + running_cost + failure_cost). The running cost charges consumption and the failure cost charges trips out of the operating envelope. The reward is never positive, and its scale differs by orders of magnitude between plants. Some plants are priced in dollars or euros and the rest are dimensionless; the per-plant summary says which. See Reward shaping for how each cost is built.

Starting state

reset samples the initial condition and the target; state has 11 fields.

Episode end

Termination. None by default. Leaving the operating envelope trips the plant: the step that leaves it is charged the trip cost and sets info["tripped"] and info["trip_cost"], and the plant restarts as reset_env would while the episode clock keeps running. Episodic use is optional: with continuing = False the plant is lost and terminated is raised on that step instead. High level carries water to the turbine; low level uncovers the tubes. See Reward shaping.

Truncation. After 400 steps.

Baselines

Recorded over 10 seeds of the 400-step episode (see Baseline results). The reward is a cost, so a return closer to zero is better.

controller	mean return	cost per step
PID	-6.783e+04	169.6
MPC	-8847	22.12

The MPC beats the PID on 10 of 10 seeds and does not trip the plant on any of them.

Arguments

The scored configuration, which tg.make returns. A keyword argument changes one field, as in tg.make("boiler_drum", max_steps_in_episode=...).

parameter	value	unit
delta_t	2	s
max_steps_in_episode	400	steps
reward_version	2	-
e_floor_level	0.00267	m
e_floor_pressure	0.05	bar
e_tol	0	m or bar
tracking_exponent	2	-
c_hold	1.566e+08	W
running_weight	1	-
failure_cost	440734	reward
restart_steps	7200	steps
rho_floor_tracking	1.91944	reward
rho_floor	1.91944	reward
V_t	88	m^3
V_d	40	m^3
V_r	37	m^3
V_dc	11	m^3
A_d	20	m^2
A_dc	0.355	m^2
L_r	11	m
m_metal	300000	kg
C_metal	550	J/(kg K)
k_friction	25	-
tau_sr	8	s
T_d	15	s
f_carry	0.1	-
f_cd	0.25	-
c_pw	5700	J/(kg K)
gravity	9.81	m/s^2
h_feedwater	1.037e+06	J/kg
Q_max	2.4e+08	W
q_feed_max	200	kg/s
Q_nominal	1.6e+08	W
q_steam_nominal	93.35	kg/s
p_nominal	85	bar
V_wt_nominal	58	m^3
q_steam_noise_std	8	kg/s
level_band	0.1	m
pressure_band	2	bar
level_trip	0.25	m
pressure_min	65	bar
level_precision_floor	0.001	m
pressure_precision_floor	0.05	bar
pressure_max	105	bar
fuel_weight	0	-