Research On Video Caching Algorithms In Mobile Edge Computing Networks

With the rapid proliferation of 5G networks and personal smart devices,mobile video services have experienced explosive growth,leading to an increase in the proportion of video data traffic in communication networks.Traditional cloud computing models are limited by the bandwidth capacity of return links,making it difficult to support large-scale video requests,leading to long response times and network congestion.To enhance the quality of user experience(Qo E)in video services and address the limitations of traditional cloud computing models,Mobile Edge Computing(MEC)has emerged.By caching videos in MEC servers,it is possible to reduce the bandwidth consumption of return links and the response time for user requests.However,due to the limited resources of MEC servers,designing efficient caching algorithms to fully utilize resources and meet the increasing demand for video services is crucial.Common caching algorithms include LRU(Least Recently Used),LFU(Least Frequently Used),and their variants.These algorithms are based on the fact that videos that have been recently or frequently accessed are more likely to be accessed again in the future.By prioritizing popular videos for caching,the quality of video services can be effectively improved.However,in mobile video networks,the access behavior of mobile users is usually diverse.It is not accurate enough to model the popularity of videos solely based on metrics like click counts,user preferences,and request records.Additionally,factors such as user mobility,user interactions,and dynamic network changes also affect video popularity.Moreover,short videos in mobile video networks often have regional and social attributes,which can also impact the performance of video caching algorithms.These challenges make traditional video caching algorithms face significant difficulties.The progress of deep reinforcement learning has provided new insights into tackling the challenges faced by traditional video caching algorithms.Existing deep reinforcement learning-based video caching algorithms utilize the perception capabilities of deep reinforcement learning to autonomously learn and optimize caching algorithms.However,research in this field is still in its early stages,and most proposed solutions primarily focus on video cache hit rate and minimizing round-trip link traffic,overlooking the impact of video bitrate selection on edge server resource consumption and user experience quality.This paper aims to address the limitations of existing work by conducting research on video caching algorithms in mobile video service using deep reinforcement learning methods.The main focus of this research and its innovative contributions are outlined as follows:(1)This paper proposes a video caching and update algorithm(DDPG-ABS)based on deep reinforcement learning to address issues such as unpredictable video popularity and limited knowledge of user network conditions in traditional video-on-demand services.The algorithm leverages the storage resources of edge servers and aims to improve user Quality of Experience(Qo E)and overall system gains.Firstly,based on the architecture of the mobile edge computing network,the algorithm models the mobile video service process,considering factors such as video popularity,user preferences,and video bitrate that impact the quality of mobile video services.It formulates the video caching optimization problem as a Markov Decision Process(MDP),and applies the DDPG algorithm from deep reinforcement learning.The DDPG algorithm takes the user’s video request history and network throughput information as input vectors to the neural network.The neural network extracts features from the user’s video request history and the intelligent agent outputs caching actions(whether to cache the video)based on these features.The intelligent agent adjusts the caching strategy based on the rewards obtained from the caching actions.Experimental comparisons with existing solutions demonstrate that the proposed algorithm outperforms in terms of system gains and Qo E improvement(2)This paper proposes a multi-base-station collaborative short video caching algorithm with user location prediction to enhance the hit rate of short videos and reduce the redundancy in the cache of multi-base-station cooperation.It starts by considering the user attributes such as historical base station visits,activity locations,and age,in the mobile edge computing multi-base-station collaborative network architecture,aiming to maximize the cache gain per unit capacity.The user mobility and short video local popularity are modeled,and the optimization problem of short video caching is transformed into a Markov Decision Process(MDP).The user’s historical base station visits,gender,age range,location,and activity semantic information are converted into one-hot vectors and input into the Continuous Bag-of-Words(CBOW)model for word vectorization.The word vectors are then input into the GRU network to predict the user’s location,determining the target base station with the highest prediction value and obtaining the short video’s local popularity.The short video cache state matrix serves as the input vector to the neural network in the A3 C algorithm.Convolutional Neural Networks(CNNs)are used to extract features of the short video cache state through multiple layers of one-dimensional convolutions.The softmax activation function is employed to output the probability of each caching action.The agent selects the caching action based on the probabilities and adjusts the caching strategy according to the reward obtained from the caching action.Experimental results comparing this proposed solution with existing approaches demonstrate that this solution achieves superior performance in terms of short video cache gain.