Compressor Surge

Hold the header pressure of a variable-speed compressor on a stepped setpoint while the consumers' demand ramps between levels, with the drive speed and an anti-surge recycle valve. Flow below the surge line trips the plant, and neither the reward nor any observation reads the surge margin directly, so the policy has to learn to keep it from the flows it does observe.

import target_gym as tg

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

Action space	Box((2,)), all actions in [-1, 1]
Observation space	Box((7,))
Tracked variable(s)	header pressure (kPa)
Episode length	1200 steps of 0.1 s (2 min)
Cite as	compressor_surge-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	speed command	-1	1
1	recycle command	-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[6].

#	meaning
0	header pressure (kPa)
1	m_c, compressor flow (kg/s)
2	m_d, flow delivered to the consumers (kg/s)
3	drive speed, percent of rated
4	drive speed setpoint, percent of rated
5	recycle valve opening (%)
6	target header pressure (kPa)

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. See Reward shaping.

Truncation. After 1200 steps.

Baselines

Recorded over 10 seeds of the 1200-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	-2.208e+06	1840
MPC	-1.227e+05	102.2

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("compressor_surge", max_steps_in_episode=...).

parameter	value	unit
delta_t	0.1	s
max_steps_in_episode	1200	steps
P01	101325	Pa
T01	288.15	K
R_gas	287.05	J/(kg K)
gamma	1.4	-
A_c	0.1	m^2
L_c	4	m
V_p	15	m^3
U_r	200	m/s
psi_c0	0.3	-
H	0.18	-
W	0.25	-
phi_join	0.78	-
k_d	0.12	kg/(s Pa^0.5)
dp_out	10000	Pa
k_r	0.15	kg/(s Pa^0.5)
check_width	50	Pa
N_min	0.7	-
N_max	1.05	-
drive_rate	0.03	1/s
tau_N	1	s
valve_open_rate	0.5	1/s
valve_close_rate	0.1	1/s
setpoint_block_steps	300	steps
demand_block_steps	200	steps
demand_ramp_s	5	s
demand_sigma	0.02	-
demand_theta	0.2	1/s
demand_dev_reset_clip	3	-
N_reset	1	-
x_reset	0.35	-
reset_newton_iters	6	-
soft_min_temperature	0.001	-
time_unit_seconds	1	s
reward_version	2	-
e_floor	0.0275	kPa
e_tol	0	kPa
tracking_exponent	2	-
dp_error_max	20.68	kPa
failure_cost	1.13101e+06	reward
restart_steps	9000	steps
c_hold	62270	W
running_weight	1	-
precision_floor	0.0275	kPa
rho_floor_tracking	1.40284e-05	reward
rho_floor	1.40284e-05	reward