| With the development of wireless communication network technology,the volume of wireless traffic is also increasing.At present,the distribution of wireless traffic in different time and space is not uniform to some extent,which causes the waste of network resources.At the same time,the development of new theories and technologies such as big data and machine learning also promotes the evolution of communication system towards self-organization,self-optimization and self-operation.Whether it is to improve network performance in the current network optimization process or to promote the intelligent development of wireless communication network system in the future,accurate prediction of wireless traffic plays an important role.Firstly,this thesis introduces the traditional time series modeling,multilayer perceptron and convolutional neural network in traffic prediction.Then,this thesis proposes a wireless traffic prediction scheme based on the characteristics of single base station.In this scheme,we build a neighbor-relation based dense convolutional network(NRDenseNet)to predict the wireless traffic by using the neighborhood characteristics of the base station Finally,we propose a wireless traffic prediction scheme based on city dimension.Based on the spatiotemporal correlation of wireless traffic and network characteristics,this scheme builds a space-time and network-performance based dense convolutional network(STNPDenseNet)to predict the wireless traffic in city dimension.For the wireless traffic prediction scheme based on the characteristics of single base station,compared with the long short-term memory model,the NRDenseNet model can capture the sudden changes of traffic better.The corresponding predicted results are more consistent with the real data and have smaller loss values.For the wireless traffic prediction scheme based on city dimension,we introduce the network performance characteristics into the spatial temporal dense convolutional network and propose STNPDenseNet model which captures the flow trend more accurately in the case of large-scale traffic prediction and is more suitable for the prediction of total traffic in the urban dimension.To sum up,the proposed NRDenseNet model and STNPDenseNet model can improve the accuracy of traffic prediction. |
