Restoration of multi-channel signal loss using autoencoder with recursive input strategy

Area of Science:

• Data Science
• Machine Learning
• Sensor Networks

Background:

• Multi-channel sensor data frequently experiences missing or corrupted values.
• These data integrity issues hinder the performance of intelligent systems.
• Current restoration methods struggle with complex correlations during random, continuous data loss.

Purpose of the Study:

• To develop an advanced data recovery algorithm for sensor data.
• To enhance the accuracy and efficiency of restoring corrupted sensor readings.
• To address limitations of existing methods in handling complex data loss patterns.

Main Methods:

• An autoencoder-based algorithm utilizing recursive feedback of reconstructed outputs.
• A dynamic termination criterion to optimize iterative refinement.
• Testing on diverse multivariate sensor datasets with various missing data scenarios.

Main Results:

• The recursive autoencoder significantly outperforms single-step autoencoder restoration.
• The method demonstrates robust performance across different datasets and missing data patterns.
• Achieved enhanced restoration accuracy and computational efficiency.

Conclusions:

• The proposed recursive autoencoder offers a scalable and adaptable solution for sensor data integrity.
• Improved data reliability enables enhanced operational efficiency in industrial applications.
• This approach is crucial for robust performance in complex sensor network environments.