Accelerated iterative identification of nonlinear systems based on the momentum approach

Area of Science:

• Control Systems Engineering
• Signal Processing
• Nonlinear System Identification

Background:

• Wiener nonlinear output-error systems present challenges due to parameter coupling between linear and nonlinear components.
• Accurate identification of these systems is crucial for control and analysis.

Purpose of the Study:

• To develop an efficient algorithm for identifying Wiener nonlinear output-error systems.
• To address and resolve the parameter coupling problem inherent in these systems.
• To improve the convergence rate and reduce computational complexity compared to existing methods.

Main Methods:

• A decomposition-based gradient iterative algorithm is proposed.
• The auxiliary model identification concept and hierarchical identification principle are applied.
• A momentum approach is incorporated to accelerate parameter estimation.

Main Results:

• The proposed algorithm effectively decomposes the Wiener system into identifiable subsystems.
• The addition of a momentum term significantly speeds up gradient descent and enhances convergence.
• Numerical simulations confirm the algorithm's superiority over traditional gradient-based methods.

Conclusions:

• The decomposition-based gradient iterative algorithm offers a computationally efficient and accurate solution for Wiener nonlinear output-error system identification.
• The algorithm demonstrates practical applicability, as shown in a crystallization process concentration determination example.