Multiple datasets containing sensor and actuator readings from industrial water treatment and distribution testbeds, recorded under both normal and attack scenarios.
SWaT Dataset (Secure Water Treatment)
The SWaT dataset was collected from a scaled-down industrial water treatment testbed designed to mimic real-world operations. It contains sensor and actuator data recorded during both normal operations and under a variety of cyberattacks. The dataset is commonly used for research in anomaly detection, cybersecurity, and industrial control systems.
WADI Dataset (Water Distribution Testbed)
The WADI dataset comes from a water distribution system testbed that simulates urban water supply networks. It includes time-series data from sensors and actuators under both normal and attack scenarios. The dataset is useful for evaluating cybersecurity methods in critical infrastructure environments.
Use case:
Cybersecurity and Anomaly Detection in Industrial Water Systems
| Example Observation (≈ 51 continuous sensors) | • level_LIT101 … LIT301 (water-level in 6 tanks) • flow_FIT101 … FIT401 (inlet / recycle flows) • pressure_PIT301, PIT501, … • conductivity_CIT401, pH_PIT302, turbidity_AIT504, etc. • binary alarm flags (high-level, low-flow) • timestamp features (hour-of-day, weekday) |
| Example Action (≈ 41 discrete/continuous actuators) | MV101, MV201 … MV504 (motorised valves: open / closed / %-open) • P101, P203, P501 (pump start/stop) • P402_VSD (variable-speed drive; 0–100 %) • dosing_P205 (NaOCl pump rate 0–100 %) • UV401 (UV lamp on/off) |
| Example Reward (higher is better; tuneable weights) | r_t = − (Energy_kWh_t × c₁) − (ChemCost_t × c₂) − ∑_i Penalty_i × 1{sensor_i outside spec} − AlarmPenalty × 1{any alarm triggered} |
| Example Transition estimate | • Logging frequency: 1 sample / s • Total duration: 11 days normal + 4 days attack ≈ 15 days • # transitions ≈ 15 days × 86 400 s/day ≈ 1 300 000 state-action-reward tuples (≈ 950 000 normal, 350 000 attack/anomaly – the latter can be filtered or kept for robustness training). |