Nonlinear model predictive control based on collective neurodynamic optimization.

Area of Science:

• Control Systems Engineering
• Artificial Intelligence
• Optimization Theory

Background:

• Nonlinear Model Predictive Control (NMPC) typically involves solving complex global optimization problems with non-convex functions or constraints.
• Existing NMPC methods often rely on linearization, which can lead to suboptimal solutions or convergence issues.

Purpose of the Study:

• To present a novel collective neurodynamic optimization approach for NMPC that does not require linearization.
• To develop a method that effectively handles non-convex cost functions and constraints in NMPC.
• To enhance the global search capabilities for NMPC problems.

Main Methods:

• Utilizes a group of Recurrent Neural Networks (RNNs) to emulate a brainstorming process for global optimization.
• Each RNN performs a constrained local search, guaranteeing convergence to a candidate solution.
• Integrates RNNs within a Particle Swarm Optimization framework to iteratively refine solutions and improve global search efficiency.

Main Results:

• The collective neurodynamic approach successfully finds global optimal solutions for NMPC problems without linearization.
• Demonstrates effective integration of global search and precise local search capabilities.
• Simulation results across various cases validate the approach's effectiveness and characteristics.

Conclusions:

• The proposed collective neurodynamic optimization approach offers a powerful, non-linear alternative for solving NMPC problems.
• This method enhances the ability to find global optima in complex control systems.
• The integration of RNNs and swarm intelligence provides a robust framework for advanced NMPC applications.