engleski

Optimal Feedback Control of Nonlinear Systems with Control Vector Constraints

This paper presents the derivation of the learning algorithm for feedback one-layer neural network controller for optimal control of nonlinear multivariable systems with control vector constraint. Derived is the off-line algorithm for the feedback optimal control problem with fixed terminal time and the algorithm for the time optimal control problem. Derivation of the algorithm is based on backpropagation through time (BPTT) algorithm, which is used as a learning algorithm for dynamic neural networks, and it is not based on the calculus of variations. The derived algorithms are used for feedback minimum-time control of the robot with two degree of freedom. The paper also considers the possibilities of on-line optimal regulation.

feedback optimal control; minimum-time control; neural networks; robot control

nije evidentirano