The Research And Application Of Reinforcement Learning Based On Support Vector Machine(SVM)

Support Vector Machine (SVM) is a class of machine learning method based on Statistical Learning Theory, which has the virtue of unique and globally optimal solution and good generalization ability. Reinforcement learning is an unsupervised learning technology and the agent can find optimal policy and perform on-line, which is recognized as an ideal technology to construct intelligent agent. But the standard Q learning algorithm is not used in the learning problem of continuous state space and continuous action space. The main work of the thesis is as follows:Firstly, the method of Support Vector Regression (SVR) is introduced in the problem of ball-interception in RoboCup, which predicts the ball's moved distance when the agent successfully intercepts the ball through the prediction model based on the collected samples. And the thesis also studies the selection of the parameters in order to predict more precisely. Then we compare the prediction model with the Artificial Neural Networks, the result shows that this model surpasses the Artificial Neural Networks in the precision of prediction.Secondly, a new reinforcement systems based on Support Vector Machine(SVM) classification is proposed, whose principle idea is to participate a few of ranges based on the distance between the present state and the goal state, and build the connection between action and state using SVM. Experiments show that the method learns the optimal policy and generalizes from the limited example space using SVM. Thus it can effectively solve the continuous state representation.