Improved WOA-DBSCAN Online Clustering Algorithm for Radar Signal Data Streams.

Area of Science:

• Signal Processing
• Data Mining
• Machine Learning

Background:

• Traditional density clustering algorithms struggle with pulsed data streams from multiple sources, leading to poor clustering and high computational costs.
• Existing methods require manual parameter tuning and re-clustering for new data, limiting efficiency.
• Aliasing in pulsed data streams poses challenges for accurate signal source identification.

Purpose of the Study:

• To develop an efficient online density clustering algorithm for pulsed data streams.
• To improve clustering accuracy and reduce computational complexity compared to traditional methods.
• To address the limitations of existing density clustering algorithms in handling aliased signals.

Main Methods:

• Proposed an online density clustering algorithm incorporating an improved golden sine whale optimization.
• Modified the neighborhood definition from a single parameter (Eps) to a joint decision (Eps and θ) for flexible range delimitation.
• Integrated flow clustering to classify pulses into valid, outlier, or inactive libraries based on cluster center distances.

Main Results:

• The proposed algorithm significantly improved sorting accuracy by 57.7% over the standard DBSCAN algorithm.
• Achieved a 37.8% improvement in sorting accuracy compared to the Whale Optimization Algorithm-DBSCAN (WOA-DBSCAN) algorithm.
• Demonstrated reduced computational load through efficient pulse analysis and classification.

Conclusions:

• The improved online density clustering algorithm effectively handles pulsed data streams with aliasing.
• The algorithm offers enhanced clustering performance, reduced computational requirements, and superior accuracy.
• This approach provides a robust solution for signal source identification in complex data environments.