Multiple actor-critic structures for continuous-time optimal control using input-output data.

Area of Science:

• Control Engineering
• Artificial Intelligence
• Nonlinear System Dynamics

Background:

• Industrial process control often involves multiple, data-dependent performance objectives.
• Existing methods may struggle with unknown nonlinear systems and varied performance criteria.

Purpose of the Study:

• To develop an optimal control strategy for unknown nonlinear systems using input-output data.
• To address situations with multiple performance objectives based on data characteristics.

Main Methods:

• Utilizing shunting inhibitory artificial neural networks (SIANN) for input-output data classification.
• Employing approximate dynamic programming with model, critic, and action networks for optimal control within categories.
• Reconstructing unknown system dynamics using a recurrent neural network (RNN) model.
• Approximating critic and action networks with neural networks (NNs).

Main Results:

• The proposed scheme achieves optimal control for unknown nonlinear systems.
• Model error and the closed unknown system are proven to be uniformly ultimately bounded.
• Simulation results validate the effectiveness of the optimal control scheme.

Conclusions:

• The developed multiple actor-critic structure provides a robust solution for optimal control in complex industrial processes.
• The integration of SIANN and RNNs enables adaptive and accurate control of unknown nonlinear systems.