Multivariate linear regression model based on cross-entropy for estimating disorganisation in drone formations

Area of Science:

• Robotics and Control Systems
• Artificial Intelligence
• Networked Systems

Background:

• Static drone swarms in applications like disaster management face unpredictable intrusions.
• Intruder avoidance in formations increases energy consumption and impacts operational efficiency.
• Balancing avoidance costs with formation functionality is crucial for swarm resilience.

Purpose of the Study:

• To develop a predictive model for optimizing drone swarm formation parameters.
• To estimate the disturbance caused by intruders based on formation settings.
• To aid in selecting parameter values that minimize energy expenditure during collision avoidance.

Main Methods:

• Utilizing multivariate linear regression to model intruder disturbance.
• Employing cross-entropy as a metric to quantify disturbance.
• Generating simulation data to train and validate the regression model.

Main Results:

• The developed model explains up to 54.4% of the variability in cross-entropy (disturbance).
• The model's predictive accuracy is twice as effective as the baseline mean cross-entropy estimator.
• Provides a data-driven approach to parameter selection for drone swarm formations.

Conclusions:

• The novel model offers significant improvements in predicting and managing intruder-related disturbances in drone swarms.
• Optimized parameter selection can lead to reduced energy consumption and enhanced swarm functionality.
• This research contributes to more robust and efficient autonomous drone operations.