Predictive control of nonlinear systems based on identification by backpropagation networks

Area of Science:

• Artificial Intelligence
• Control Systems Engineering
• Computational Neuroscience

Background:

• Multilayer perceptron neural networks possess universal approximation capabilities.
• Discrete nonlinear dynamical systems present significant modeling and control challenges.
• Accurate identification of system dynamics is crucial for effective control.

Purpose of the Study:

• To model discrete nonlinear dynamical systems using a two-hidden-layer perceptron.
• To develop a one-step-ahead predictive control scheme based on the identified model.
• To enhance the predictive control scheme into an adaptive version robust to identification uncertainties.

Main Methods:

• Utilizing the universal approximation property of multilayer perceptron neural networks.
• Employing a backpropagation algorithm for training the neural network model.
• Implementing a nonlinear optimization process for determining future control inputs.
• Leveraging online learning capabilities for adaptive control.

Main Results:

• The neural network model accurately identifies nonlinear dynamical systems.
• The proposed one-step-ahead predictive control scheme effectively manages the identified systems.
• The adaptive control scheme demonstrates robustness against incomplete system identification.
• Simulation results confirm the efficacy of the neural control scheme for complex nonlinear systems.

Conclusions:

• Neural networks provide a powerful tool for modeling and controlling discrete nonlinear dynamical systems.
• The developed adaptive predictive control strategy offers a robust solution for complex control problems.
• This approach is effective even when faced with uncertainties in system identification.