- Anomaly detection
Anomaly detection technology for plants: “2-stage auto encoder”
Detects anomaly signs earlier than human detection, by finding unnoticeable changes in complex, large-scale plants.
- Detects anomalies from a large amount of monitoring sensor signals at power plants.
- Detects anomalies of power plants, even in changing conditions where sensor signals accodingly changes with the conditions.
- Power plants
- Water treatment plants
Benchmarks, strengths, and track record
- Achieved an F1 score of 0.777 in a public data set of water treatment plants. (WADI) (0.695 with conventional technologies)
*WADI: Water Distribution; sensor signal data from a scaled-down testbed of a water treatment plant. Anomalies are caused based on 15 scenarios to 122 sensors.
*F1 score: A metric of a model’s accuracy, indicating performance of both how many of true anomalies the model discovers and how precisely the model detects anomalies.
- Currently conducting experimental operation of anomaly detection service at the Mikawa Power Plant operated by SIGMA POWER Ariake Corporation, a subsidiary of Toshiba Energy Systems & Solutions Corporation
Please include the title “Toshiba AI Technology Catalog: ‘2-stage auto encoder’ plant anomaly sign detection technology” or the URL in the inquiry text.
Please note that because this technology is currently the subject of R&D activities, immediate responses to inquiries may not be possible.
- Development of anomaly sign detection AI that quickly and accurately detects anomalies buried in changes in the status of large-scale, complex plants (press release; December 7, 2021)
- S. Naito, Y. Taguchi, K. Nakata, Y. Kato, “Anomaly Detection for Multivariate Time Series on Large-scale Fluid Handling Plant Using Two-stage Autoencoder,” 2nd Workshop on Large-scale Industrial Time Series Analysis, hosted by 21st IEEE International Conference on Data Mining (ICDM2021 LITSA), 2021.
- S. Naito, Y. Taguchi, Y. Kato, K. Nakata, R. Miyake, I. Nagura, S Tominaga, T Aoki, “Anomaly sign detection by monitoring thousands of process values using a two-stage autoencoder,” Mechanical Engineering Journal, Volume 8, Issue 4, 2021.