Spatial Gaussian process regression with mobile sensor networks

Area of Science:

• Computer Science
• Electrical Engineering
• Robotics

Background:

• Mobile wireless sensor networks (MWSNs) require efficient methods for modeling spatial and spatiotemporal functions.
• Existing methods may face challenges with distributed computation and adaptability in dynamic environments.

Purpose of the Study:

• To develop a distributed Gaussian process regression (DGPR) approach for MWSNs.
• To enable independent regression computations at each sensor node.
• To facilitate adaptive modeling of spatiotemporal functions in mobile networks.

Main Methods:

• Utilized sparse Gaussian process regression with a compactly supported covariance function.
• Developed a neighbor-to-neighbor communication protocol for distributed computation.
• Integrated an information entropy-based locational optimization algorithm for collective motion control.

Main Results:

• The DGPR approach allows independent regression results from each sensor node.
• The method effectively models both stationary spatial functions and dynamic spatiotemporal functions.
• Simulations demonstrated the approach's performance and adaptability.

Conclusions:

• The proposed DGPR method offers an efficient and scalable solution for MWSNs.
• The approach enhances the network's ability to adapt to changing environments and functions.
• DGPR facilitates intelligent collective motion control through distributed processing.