Lattice dynamical wavelet neural networks implemented using particle swarm optimization for spatio-temporal system...

Area of Science:

• Artificial Intelligence
• Computational Science
• Dynamical Systems

Background:

• Spatio-temporal system identification is crucial for understanding complex dynamic processes.
• Existing adaptive wavelet neural networks may lack efficiency and parsimony.
• Coupled map lattice models offer a framework for simulating spatio-temporal dynamics.

Purpose of the Study:

• Introduce a novel family of adaptive wavelet neural networks, Lattice Dynamical Wavelet Neural Networks (LDWNNs).
• Develop an efficient and parsimonious modeling framework for spatio-temporal system identification.
• Enhance network training using a hybrid approach combining optimization and refinement algorithms.

Main Methods:

• Combined an efficient wavelet representation with a coupled map lattice model.
• Proposed a new orthogonal projection pursuit (OPP) method integrated with particle swarm optimization (PSO).
• Developed a two-stage hybrid training scheme: adaptive neuron recruitment and redundant neuron removal.

Main Results:

• Successfully introduced LDWNNs for spatio-temporal system identification.
• Demonstrated the effectiveness of the OPP-PSO algorithm for network augmentation.
• Validated the two-stage training scheme in constructing parsimonious network models.
• Presented a real-world spatio-temporal system identification example showcasing the framework's performance.

Conclusions:

• The proposed LDWNNs offer an efficient and adaptive approach to spatio-temporal system identification.
• The hybrid training scheme ensures parsimony and accuracy in the developed models.
• The framework provides a robust solution for complex spatio-temporal modeling challenges.