Load Swing Suppression of Crane System by Neuro-Controller Utilizing GA with Rough Evaluation

摘要

The present paper proposes a new design method in which a controller acquires the required performance automatically. Two important tasks of the control designer are the creation of a simulator of the plant dynamics and the determination of the evaluation of the controller. A three layered neural network of 4-inputs and l-output controls the plant. The proposed method requires little knowledge of difficult control theories. Genetic algorithm (GA) evolves the neural networks to enable the control of nonlinear systems. The evaluation method originally proposed in this paper is very simple and easy to understand, so that it can be used even by non-professionals working in the field of control. In order to demonstrate how to apply the proposed method, the present paper analyzes the load swing suppression of a crane system on a cart. Simulations reveal that the neurocontroller has higher performance than the linear quadratic regulator (LQR) with respect t settling time and robustness. The neuro-controller is able to suppress load swing even in environ- ments that exceed design specifications.