Research And Implementation Of Base Station Traffic Prediction Algorithm Based On Spatio-Temporal Characteristics

With the development of society and science and technology,mobile communication has become an important thing to support daily life,and the construction of communication base station as the basic unit carrying mobile communication capabilities is crucial.In recent years,the field of mobile communication has developed rapidly,the global demand for services has surged,and the scale of mobile devices and the Internet of Things has shown exponential growth,which has brought opportunities and challenges to the development of mobile communications.On the one hand,the high liquidity of demand makes it necessary to support a large amount of infrastructure,on the other hand,resource redundancy makes mobile communication costs high and limits development,which makes base station traffic prediction a key issue.Accurate and efficient base station traffic prediction can have a timely and effective grasp of the business situation of the base station in the future period,so as to expand and shrink the base station so that it can meet the demand without causing a lot of waste.At the beginning of the research,the focus was on the independent traffic size of each base station,with the aim of adjusting the base station capacity separately.As spatiotemporal interaction becomes easier,the research problem is no longer limited to traffic size,but instead extends to where traffic comes from and is affected.As a whole,the interaction dynamic information of the base station group becomes increasingly important.However,the spatiotemporal interaction,highdimensional complexity,mutation and other characteristics of the base station network itself bring great difficulties to traffic prediction.On the one hand,the time component of independent base station traffic value prediction is complex,spatial feature extraction is difficult,and contingency is large;On the other hand,the traffic trend data of base station groups has complex dimensions and sparse high-dimensional data.Therefore,in order to reasonably model these characteristics of the base station network and complete the base station traffic prediction more accurately and efficiently,this paper studies the problem of base station network traffic prediction,and discusses and models the two problems of traffic value and trend,the main work is as follows:(1)A traffic prediction model based on sequence component decomposition and spatiotemporal feature modeling is proposed.This model consists of component decomposition and spatiotemporal models.The former completes the decomposition of time series components through the EEMD(set empirical mode decomposition)method and models the obvious periodic law.The latter performs spatiotemporal modeling for a single regular component,the same spatiotemporal model models model spatial characteristics,and different spatiotemporal models superimpose temporal trends.A spatiotemporal model consists of spatiotemporal modeling and sequence prediction.Spatiotemporal modeling uses multiple convolutional neural networks to complete spatiotemporal feature extraction,multi-channel input fuses temporal and spatial features to eliminate temporal contingency,and CNN without pooling layer combines inception and extended convolution to reduce model complexity and accelerate iteration.The LSTM model is used to receive the spatiotemporal features of the output of the convolutional neural network and complete the result prediction.(2)A traffic trend prediction model based on high-dimensional modeling and vector mapping is proposed.The model consists of a highdimensional transformation trend module and a prediction module.The high-dimensional transformation trend module is used to solve the problem that the trend data information is complex and difficult to express.Firstly,aiming at the problem of inconsistent expression of flow and flow direction,the flow-flow matrix is transformed into multiple flow-geographic distribution matrices,so that the flow direction is expressed by the flow value,and then the flow direction information is modeled by the concept of image depth.Secondly,according to the proximity and large-scale characteristics of the base station network itself,the Embedding layer is used to map the data in low dimensions to reduce the sparsity.The highdimensional conversion trend module realizes the transformation of traffic direction to value,and completes the unification of model prediction targets.Then,the prediction of the flow value is carried out through the prediction module to complete the final result prediction.