A nonlinear control method based on ANFIS and multiple models for a class of SISO nonlinear systems and its...

Area of Science:

• Control Systems Engineering
• Artificial Intelligence
• Nonlinear Dynamics

Background:

• Uncertain discrete-time nonlinear systems with unstable zero-dynamics pose significant control challenges.
• Existing control strategies often rely on restrictive assumptions regarding unmodeled dynamics, limiting their applicability.
• Improving closed-loop stability and dynamic performance in such systems remains an active research area.

Purpose of the Study:

• To develop a novel nonlinear control strategy for uncertain single-input, single-output discrete-time nonlinear systems.
• To enhance system robustness and dynamic performance by integrating adaptive-network-based fuzzy inference systems (ANFIS) with multiple models.
• To relax common assumptions on unmodeled dynamics and improve learning convergence rates.

Main Methods:

• A hybrid control architecture combining a linear robust controller, an ANFIS-based nonlinear controller, and a switching mechanism.
• Integration of these components using a multiple models technique for uncertain systems.
• Application of a "one-to-one mapping" technique to ensure the universal approximation property of ANFIS.

Main Results:

• The linear controller ensures boundedness of input and output signals.
• The ANFIS-based nonlinear controller improves the dynamic performance of the closed-loop system.
• The switching mechanism guarantees simultaneous closed-loop stability and performance enhancement.
• The proposed method relaxes uniform boundedness assumptions on unmodeled dynamics, increasing applicability.
• ANFIS-based compensation of unmodeled dynamics accelerates neural network learning convergence.
• Effectiveness validated through a numerical example and an alumina sintering process.

Conclusions:

• The novel nonlinear control strategy effectively addresses uncertain discrete-time nonlinear systems with unstable zero-dynamics.
• The integration of ANFIS, multiple models, and a switching mechanism offers significant advantages in stability, performance, and applicability.
• The controller demonstrates enhanced robustness and faster learning convergence compared to existing methods.