A Complex Varying-Parameter Convergent-Differential Neural-Network for Solving Online Time-Varying Complex Sylvester...

Area of Science:

• Computational neuroscience
• Applied mathematics
• Artificial intelligence

Background:

• Solving time-varying complex Sylvester equations is crucial in various scientific and engineering fields.
• Existing methods, such as complex fixed-parameter convergent-differential neural networks (CFP-CDNN), face limitations in convergence speed and adaptability.
• Recurrent neural networks offer a promising approach for dynamic system modeling and equation solving.

Purpose of the Study:

• To propose a novel recurrent neural network, the complex varying-parameter convergent-differential neural network (CVP-CDNN), for solving time-varying complex Sylvester equations.
• To investigate the effectiveness and performance of two types of CVP-CDNNs (Type I and Type II).
• To explore the impact of different activation functions on the convergence properties of the CVP-CDNN.

Main Methods:

• Development and theoretical analysis of two novel CVP-CDNN architectures.
• Mathematical proofs demonstrating the convergence properties of the proposed networks.
• Comparative analysis using computer simulations against traditional complex fixed-parameter convergent-differential neural networks (CFP-CDNN).
• Evaluation of CVP-CDNN performance with various activation functions, including linear and sign-bi-power.

Main Results:

• The proposed CVP-CDNNs (Type I and Type II) are proven effective for solving time-varying complex Sylvester equations.
• Super-exponential convergence performance is achieved when using a linear activation function.
• The CVP-CDNN with a sign-bi-power activation function exhibits finite-time convergence.
• The convergence time of the CVP-CDNN with sign-bi-power activation is significantly shorter than that of the CFP-CDNN.
• Computer simulations confirm superior convergence performance of the CVP-CDNN over the CFP-CDNN.

Conclusions:

• The novel CVP-CDNN represents a significant advancement in solving time-varying complex Sylvester equations.
• The CVP-CDNN offers enhanced convergence speed and performance compared to existing CFP-CDNN methods.
• The choice of activation function critically influences the convergence characteristics, with sign-bi-power offering finite-time convergence benefits.