Grid index subspace constructed locally weighted learning identification modeling for high dimensional ship...

Area of Science:

• Computational Science
• Machine Learning
• Marine Engineering

Background:

• Off-line locally weighted learning (LWL) requires storing all training data, causing significant computational load with large datasets.
• High dimensionality in ship maneuvering systems exacerbates these computational challenges.

Purpose of the Study:

• To develop an efficient algorithm for locally weighted learning (LWL) in high-dimensional ship maneuvering systems.
• To reduce the computational burden and ensure real-time performance for LWL applications.

Main Methods:

• A grid index subspace constructed algorithm is proposed.
• High-dimensional training data is encoded and stored in an equal-interval grid.
• Query points are encoded to identify their corresponding grid, enabling subspace allocation for reduced computation.

Main Results:

• The proposed algorithm significantly reduces computational complexity by performing LWL within allocated subspaces.
• This approach avoids the need for neighborhood searches common in clustering algorithms.
• Theoretical calculations and simulations validate the effectiveness and efficiency of the scheme.

Conclusions:

• The grid index subspace algorithm effectively addresses the computational burden of LWL in high-dimensional data.
• The method ensures real-time performance, making it suitable for complex systems like ship maneuvering.
• This provides a scalable and efficient solution for data-intensive learning tasks.