A Class Of Intelligent Control Method Based On Learning Human Strategy And Its Stability Analysis

In these years, controller design for complex dynamic systems becomes increasingly animportant and challenging topic. Many research achievements have been gotten in this area.An intelligent control method based on learning human strategy is proposed in this paper. Itbuilds the model of the relationship of the system state and the human’s control input. Wecollect the information of the state of the system and the human control. Then we get thelearning controller by training through these samples by Support Vector Machine (SVM).As different samples are trained to different controllers, we want to analyze the controllerand select the controller whose performance is better before putting it into practice. Thecontroller trained by SVM is a typical nonlinear system. First a necessary and sufficientcondition is given to prove the stability of the system under perturbations. For the system thatis locally stable, the size of domain of attraction (DOA) reflects the immunity of the system.In this paper, Sum of Squares (SOS) method is introduced to estimate the DOA. To solve thebilinear problem encountered during using SOS to estimate the DOA, two iterated algorithmsare proposed and an example confirms the correctness of these algorithms.In order to verify the above control and stability analysis method, these methods areapplied in a single wheel robot called Gyrover to make the robot realize self-balancing.Through collecting the data of sensors when human expert control the robot, trained to get thelearning controller. The simulation experiment results verify the effectiveness of the learninghuman strategy method. Then we analyze the controllers got by training. The simulationresults show that the system has the stronger immunity to disturbance with the larger DOA.