The HAI dataset was collected from a realistic industrial control system (ICS) testbed augmented with a Hardware-In-the-Loop (HIL) simulator that emulates steam-turbine power generation and pumped-storage hydropower generation.
Use case
Optimal Control of a Pumped Hydro Storage Plant
| Example observations | 59 analog sensors: reservoir level, bearing temperature, flow rate, upstream/downstream pressure, active power, bus voltage, etc. 11 binary sensors: switch status, alarms, etc. 9 network/PLC measurements: latency, error rate, etc. Timestamp |
| Example actions | Valve VF101 opening (0–100%) Pump P201 frequency setpoint (RPM) Guide vane position of turbine TV301 Start/Stop of drainage pump, etc. |
| Example reward | +c_1*P_prop – c_2*E_pump – c_3*alarm – c_4*|a_t| Where: P_prod (in kW) E_pump = pump_speed*current (power consumption of the pump) alarm = 1 if any fault tag or critical range is detected (already labeled in HAI) a_t penalizes abrupt control changes Coefficients c_1..c_4 are adjustable, e.g., (1, 1, 10, 0.1) |