Adaptive Bitrate Optimization Algorithm Based On Reinforcement Learning And Its Application In Water Quality Monitoring

According to the forecast of the Cisco Global Mobile Data Traffic White Paper(2016-2021),video traffic will account for three quarters of global mobile data traffic in 2021.To cope with the explosive growth of video data,the international standard organization MPEG,in collaboration with major technology companies,has developed the HTTP Adaptive Streaming(HAS)technology's international standard(DASH),also known as MPEG-DASH standard.The MPEG-DASH standard stipulates specific technical parameters for audio and video content formats,transmission methods,and service control,but opens up the definition of acquisition strategies for client-side video clips.Therefore,research on adaptive rate algorithms has great application value for video data playback according to MPEG-DASH standards.The existing rate adaptation algorithms are mostly based on the heuristic method.This method generally implements the adaptation of the code rate by hard coding according to the specific network conditions.This leads to a rate adaptive algorithm based on this method can not effectively respond to changes in the network environment.In addition,there are also many scholars who use the reinforcement learning algorithm to cope with the problem of adaptive rate adaptation under the changing network environment.These algorithms often use the Q learning algorithm in reinforcement learning to learn the optimized code rate adaptation strategy.Although compared with the heuristic method,Q learning can obtain high quality of user experience,but Q learning has the disadvantages of difficult to encode continuous state values and slow learning convergence in large state space.For this reason,this paper combines the nearest neighbor algorithm and designs a Q learning algorithm that can handle the continuous state.In addition,aiming at the shortcoming of slow convergence of Q learning,combined with the theory of brain coding on neurons,a population-encoding Q learning algorithm is proposed..The main work of the thesis is summarized as follows:1.In this paper,we combine the nearest neighbor algorithm to design a Q learning algorithm that can handle states with continuous values.Experimental results show that the rate optimization algorithm based on nearest neighbor Q learning has higher user experience quality.2.Aiming at the shortcoming of the slow convergence of Q learning in the large state space,combined with the brain's theory of neuron population coding,a group coding Q learning algorithm is proposed.3.Developed a video data management platform for river basin water quality monitoring,and achieved the results of real-time monitoring of the whole basin,centralized centralized management,and coordinated monitoring and early warning between multiple monitoring points.