Research On Adaptive Bitrate Video Streaming Technology Based On Broad Reinforcement Learning

In recent years,with the rapid development of Internet communication technology and smart terminals,the proportion of mobile video traffic in global data traffic has continued to rise,and users' quality requirements for streaming media service have become higher and higher.However,highquality video transmission will take up too much wireless network bandwidth.Dynamic Adaptive Streaming over HTTP(DASH)can adaptively switch bitrate of media segment according to the external environment,providing users with high-quality and smooth viewing experience under limited network resources.However,the current DASH approach still has two shortcomings.For one thing,when measuring users' viewing experience,it only considers quality of service(Qo S)related to network parameters,and cannot provide users with a personalized video experience.For another,it is difficult for its bitrate selection algorithm to achieve the global optimal solution through traditional optimization methods,and its performance is often limited in actual scenarios.Therefore,this thesis proposes an adaptive bitrate video streaming technology based on broad reinforcement learning to implement an adaptive video streaming system.The main research contents of this thesis are as follows:Firstly,an evaluation model based on Quality of Experience(QoE)is construct.On the one hand,from the aspects of QoE influencing factors,the relationship between multiple influencing factors and user experience is defined and qualitatively analyzed.On the other hand,considering the feature importance,through the three indicators of Pearson's correlation coefficient,information gain,and the importance of random forest features to select factors that have a greater impact on the quality of user experience.According to the technical characteristics of DASH,a multi-dimensional linear QoE evaluation model is established to achieve an accurate evaluation of user experience quality,in which the least square method is used to calculate the parameters.Secondly,an adaptive bitrate video streaming technology based on broad reinforcement learning(BRL)is proposed.In order to solve the problems of deep reinforcement learning(DRL)for high computing power,high model complexity and long training time,the BRL general model was first proposed.It contains broad learning system instead of traditional deep neural network,because broad learning system can quickly calculate and dynamically adjust the network structure.Therefore,the BRL approach not only has the extended learning ability of reinforcement learning,but also can be trained quickly.Then,an adaptive bitrate video streaming technology based on BRL is proposed.According to the optimization decision of reinforcement learning,the code rate adaptation process is modeled.The state space is defined as the state characteristics that can fully reflect the environmental changes during the playback process of the client.The users' experience quality obtained by the QoE evaluation model is used as the reward to guide the proposed adaptive bitrate algorithm to learn and update.Through these suitable settings,the performance of the proposed approach has been improved.Finally,this thesis constructs an adaptive video streaming system based on DASH,and evaluates the proposed algorithm.The experimental results show that compared with the fixed strategy rate adaptive algorithm,the adaptive bitrate video streaming technology based on broad reinforcement learning proposed in this thesis has better scalability.It can provide users with a better viewing experience in a complex environment.At the same time,by adjusting the penalty coefficient of the reward function,the proposed algorithm can provide more targeted options for different users.Compared with DRL approach,the training speed of the proposed BRL approach in this thesis is increased by 53.9%,which can learn and response faster.