Load Reduction Model Based On Stream Data

With increasing demand on wireless services in 5G,equipment supporting multimedia applications has been becoming more and more popular in recent years.With billions of devices involved in mobile Internet,data volume is undergoing an extremely rapid growth.As a huge number of networking and computing equipment which generate big data,integrated into the 5G system,energy efficiency becomes another major challenge in building a green 5G system.Therefore,data processing,network overload,and energy consumption have become three urgent problems.In this thesis,we attempt to solve these three problems from the perspective of load reduction.Thus,we propose two models,i.e.,hybrid-stream big data analytics model and big data computing architecture.The first model is applied to perform stream data like video surveillance,and the second model is contributed to network energy consumption from the aspect of feature.What's more,two strategies,i.e.,feature reduction strategy(FRS)and rule certainty update strategy(RCUS),are proposed to achieve feature reduction.FRS aims to eliminate redundant system monitoring information by judging correlation among features,and RCUS mines both positive and negative rules to help reduce load in network.Simulation results show that our methods(both model and strategy)are efficient and robust enough to keep up with the big data crush in 5G era.The main contributions in this thesis can be summarized as follows:(1)Through the analysis of correlation of multimedia big data,this thesis proposes a hybrid-stream big data analytics model.The model adds time dimension to original convolutional neural network.It aims to obtain the reduction of other dimensions at the expense of adding time dimension in computing process.Experimental results show that redundant video data can be removed by labeling the video frames or segments with information values,which can reduce the load of multimedia data stream.(2)Considering the problem of long training time and huge memory resources for feature extraction of streaming data in the network,this thesis proposes a big data computing architecture.Besides,a feature reduction strategy(FRS)based on this architecture is also proposed.The strategy uses feature grouping to speed up the training speed of sample data and uses adaptive scale variables to reduce feature selection bias.Experimental results show that this strategy can effectively extract data features under the premise of fast convergence.(3)Due to the problem of computing association rules' increasing complexity with negative features,this thesis proposes a rule certainty update strategy(RCUS).On the basis of improved evolutionary algorithm,this strategy introduces gene expression to digitize mining rules and combines certainty condition to ensure the rationality of mining rules.Experimental results show that this strategy can improve the efficiency and the final accuracy of positive and negative mining rules.