Research On Community Mobile Data Traffic Prediction Algorithm Based On Machine Learning

As one of the basic technologies for intelligent management and scheduling of mobile networks,mobile data traffic prediction has been widely used in mobile communication network wireless resource allocation and scheduling strategies,and assisted base station site selection and construction in recent years.Accurate prediction of community mobile traffic can help telecom operators to allocate network resources reasonably,improve the stability of mobile networks,especially to meet the new needs of future 5g technology,save 5g base station construction and operation costs for operators,and promote network management The intelligent transformation of the center is of great significance.In the original communication data flow sequence collected by the mobile base station,due to missing data sampling,wrong sampling or abnormal emergencies,there may be data missing,sudden increase points,sudden drops,etc.,and future traffic forecasts based on abnormal data Accuracy has a greater impact.Based on the rules of the voting method,this article uses the ring comparison method,the year-on-year detection method,and the LOF detection method for comprehensive inspection and screening of abnormal points.The cubic spline interpolation method is used to smooth the selected abnormal points..Then,according to the processed communication data,the time dimension characteristic analysis of the mobile communication data flow sequence and the spatial correlation analysis of the communication flow sequence of adjacent communities are carried out.Aiming at the problem that the traditional time series forecasting model is easily affected by the data noise existing in the original time series,this paper first uses the EMD algorithm to decompose and reduce the noise of the traffic data time series,and then constructs a community mobile data traffic forecast based on LSTM model.The verification experiment of the above method is carried out on the actual collected data set.The results show that the use of EMD algorithm to decompose the time series after denoising and fusion with the LSTM network for prediction is helpful to improve the prediction accuracy of the original single LSTM network.The study found that there are great similarities in the mobile data traffic of adjacent communities covered by one or more base stations.Therefore,if the mobile data traffic of adjacent communities is formed into a spatial dimension,then the community mobile communication traffic data sequence is It can form a spatio-temporal data sequence,thereby increasing the feature quantity of the prediction model,which is beneficial to improving the prediction accuracy.Based on the above analysis,this paper proposes a traffic prediction method based on VP-Tree and attention mechanism;using VP-Tree to construct the spatial dimension of mobile data traffic,using Seq2 Seq as the basic structure,the attention mechanism is introduced to improve mobile data traffic.The data at key time points in the sequence are given appropriate weights to further improve the prediction accuracy.In this paper,a comparison experiment between the above algorithm and the traditional method is carried out on the real data set provided by the operator.The results show that the method of noise reduction and decomposition for the community mobile communication data flow sequence in the time dimension is helpful to improve the machine learning algorithm.The prediction accuracy of,and the introduction of the spatial dimension,the VP-Tree based on the space-time dimension proposed in this paper combined with the attention mechanism of the deep learning prediction model,its prediction accuracy and versatility are better than other single-consideration community mobile communication data traffic sequences The time feature method,although the training time is slightly longer than other methods,can meet the actual application requirements.