Design and Implementation of Surakarta Game System Based on Reinforcement Learning

Based on Alpha Zero's idea, this paper designs and implements a deep convolution neural network model with situation assessment and strategy function modules using Tensorflow framework. The model generates sample data through evolutionary iteration and self-matching game, effectively resolves the missing high-quality calibration data in some rounds faced by supervised model training methods. This paper also proposes a reinforcement learning method with opponents, that is to say, first of all, taking a Surakarta chess game system with a certain level of intelligence as the opponent model to carry out "catch-up learning", which improves the learning efficiency of the target model in the training process. Experiments show that this method is effective, and the system can learn to play chess and play good chess by itself according to the rules of the game.