Research On Video Game Simulation Algorithms Based On Deep Reinforcement Learning

With the development of computer graphics technology,network technology and human-computer interaction technology,the research of video games has entered a new stage.The input picture of early video games is rough,and the difficulty of the game is single,which can not give gamers a better game experience.In order to create a more realistic game virtual environment,increase the playability and challenge of the game to achieve a more realistic interaction between players and game agents,the research of video game simulation becomes very important.For a long time,reinforcement learning has been limited in scope since it was unable to directly process raw sensory data from the environment.Deep learning is developing very fast in recent years and the features in high-dimensional images can be extracted automatically by deep neural network.Therefore,more and more scholars are beginning to study the combination of deep learning and reinforcement learning,namely deep reinforcement learning.In order to solve the problem of low score and slow learning strategy of deep reinforcement learning in video game simulation,a video game simulation algorithm based on improved Deep Q-network(DQN)is proposed.Firstly,the algorithm improves the activation function of convolutional neural network(CNN).Design and construct a segmentation activation function combining the advantages of both ReLU and Softplus activation functions.The improved activation function is used as the activation function of the full connection layer of CNN.Secondly,an improved Gabor filter is designed to replace the original trainable filter in CNN.Finally,each frame image of video game and the improved Gabor filter are convoluted to get the features in different directions,and then the features are fused,and using kernel principal component analysis(KPCA)to reduce dimensionality of the merged features,which replaces the original video game image as the input of the CNN.The Q-Learning algorithm of reinforcement learning is used to train and update the network.Using Q-Learning algorithm of reinforcement learning to train and update network weight,then the training model can be obtained and the simulation of video game can be realized..Experiments show that the enhanced learning model combined with deep neural network can successfully learn the control strategy by training with the improved activation function and the improved Gabor filter.Compared with the traditional reinforcement learning model,the improved model performs better on video games and learns the optimal strategy faster.