Robust data cleaning methodology using online support vector regression for ultra-short baseline positioning system.

Area of Science:

• Marine technology
• Acoustic positioning systems
• Data science

Background:

• Ultrashort baseline (USBL) acoustic positioning systems are crucial for underwater navigation due to their simplicity and maneuverability.
• Raw USBL data quality can be compromised by measurement outliers and missing values, impacting navigation reliability.

Purpose of the Study:

• To develop a robust and efficient data cleaning methodology for USBL acoustic positioning systems.
• To improve the accuracy and consistency of USBL navigation data in real-time applications.

Main Methods:

• A novel data cleaning approach utilizing Online Support Vector Regression (OSVR) was developed.
• Sliding-window samples were employed to train the OSVR model for online time series prediction.
• One-step-ahead predictions from OSVR were used to detect and replace outliers and supplement missing data.

Main Results:

• The proposed OSVR methodology demonstrated effective real-time navigation capabilities.
• Consecutive and consistent positioning data were acquired for the USBL system.
• Root mean square errors in longitude and latitude were significantly reduced by 91.75% and 85.53%, respectively, compared to raw data.
• The OSVR method outperformed traditional Least Square (LS) and kernel recursive LS algorithms.

Conclusions:

• The OSVR-based data cleaning methodology significantly enhances the quality of USBL acoustic positioning data.
• This approach ensures reliable and accurate underwater navigation for human-occupied vehicles.
• The proposed method offers a superior alternative to existing data cleaning techniques for USBL systems.