Controller design for discrete input control system based on machine-learning

Switching control is an effective control technique for control systems equipped with low-resolution actuators. It is modeled as a control system that restricts its input to discrete values. Designing a discrete-valued controller is equivalent to determining the switching surface. In this paper, a controller design method based on a machine learning technique is discussed. In particular, this paper proposes that the supper vector machine (SVM) is used to construct the switching condition. The relation between the current situation (previous input sequence and previous output sequence), applied input sequence, and output evolution is learned by a support vector machine (SVM) from the database of previous control results. As a result, the learned SVM can determine the most suitable input for the current situation and the reference output. The effectiveness of the proposed method is verified for a discrete input system via experiments.