CPDS-AD: Cyber-Attack Detection Data from a Physical Distribution System Testbed

25 days of normal-operation data and 5 days per attack scenario were collected from a laboratory-scale 13-node three-phase unbalanced radial distribution network. The dataset includes additive, subtractive, and replay false data injection attacks, as well as denial-of-service attacks, with processed time-series and traffic features. It was created by Yulin Liu and last updated on 2026-04-12.

Use Cases

• Training machine learning models for false data injection attack detection based on structured measurement time-series.
• Developing algorithms for denial-of-service attack identification based on aggregated Modbus traffic features over 10-second windows.
• Benchmarking cyber-attack detection methods using labeled data from a realistic physical testbed rather than pure simulation.
• Studying the impact of replay attacks on cyber-physical distribution systems using synchronized timestamp data.

Strengths

• Data originates from a physical laboratory-scale testbed with a realistic 13-node distribution network and three-layer cyber architecture.
• Training set contains 14,400 measurement time points from 25 days of normal operation.
• Each of the four attack categories in the test set contains 2,880 measurement time points from 5 days of data.
• Raw communication logs were processed through message parsing, timestamp alignment, and measurement synchronization.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment for specific model training needs.

Provenance

Source

Yulin Liu

Collection Method

Developed from a laboratory-scale cyber-physical distribution system testbed, with data collected from a 13-node three-phase unbalanced radial distribution network.

Freshness

Last updated 2026-04-12 13:00:07; freshness should be verified.