移动网络优化中业务量预测及移动用户高度分层方法研究

发布时间：2018-07-23 12:35

【摘要】：虚拟化和容器技术使得核心网资源分配越来越灵活高效。因此需要网络能够提前对业务量进行感知和预测。通过预测话务量和用户数并按照预测的结果合理分配有限的网络资源,能有效提高服务质量。另一方面,随着城市高层楼宇增多,深度覆盖特别是垂直覆盖的优化也变得越来越重要,但测试人员无法进入高层建筑测试也是这一问题的难点。寻求能有效的利用用户数据来判断高层楼宇内的信号覆盖质量的方法,在学术领域和工业界还属于空白。首先,本文介绍LTE系统核心网及接入网络的网络架构,给出通过LTE系统接口进行控制面信令探测,进而利用控制面信令及用户面数据,分析得到用户数、话务量和用户感知的Wi Fi信息的过程。本文还研究了时间序列的原理与特点,分析了利用时间序列分析中广泛使用的ARIMA的原理以及模型识别的标准,并给出了预测的步骤与预测的评价标准。其次,本文对每小时话务量的数据特性进行分析,并基于分析给出预测话务量的模型。研究了话务量预测中乘积季节ARIMA模型的应用条件,给出了模型识别、阶数确定与残差校验等建模步骤。并应用STL方法将话务量时间序列分解为季节项、趋势项以及随机项,对序列进行季节性调整后再用ETS模型拟合,预测时以最近一周期的数据作为季节项的预测结果。此外还分析了Holt-Winters加法和乘法模型,并探讨了利用BP神经网络训练建模并预测的过程。本文应用这四种模型对某运营商在特定景区的话务量及终端数量进行预测,实验表明预测精度良好,能够满足移动网络性能优化要求,通过代码实现并应用于运营商核心网的网络优化中。最后,本文在利用LTE网络用户面深度包分析获得特定区域及建筑物内用户可感知的Wi Fi物理地址及RSSI后,通过数据积累得到高层楼宇的Wi Fi信息,构建Wi Fi的能量矩阵,矩阵的元素是采样点采集到的Wi Fi的RSSI值。两个Wi Fi的相关系数是能量矩阵对应的Wi Fi列之间的相关性,进而获得Wi Fi的相关矩阵。应用k-means、PAM、谱聚类与Fast Unfolding算法对Wi Fi样本进行聚类分析,获得三个Wi Fi簇。在确定底层簇后,利用簇间相关性确定其他簇,得到Wi Fi的高度标签。最后依据LTE终端测量的RSRP的强度确定移动用户的信号覆盖质量。实验结果具备较高精度并能满足无线侧网络优化需求。
[Abstract]:Virtualization and container technology make the distribution of core network more and more flexible and efficient. Therefore, it is necessary for the network to be able to perceive and predict the volume of business in advance. By predicting the traffic volume and the number of users and rationally distributing the limited network resources according to the predicted results, the quality of service can be improved effectively. On the other hand, with the increase of the high level of urban buildings in the city The optimization of depth coverage, especially the vertical coverage, is becoming more and more important, but it is also a difficult problem for the tester to be unable to enter the high-rise building test. To find a method that can effectively use the user data to judge the quality of the signal coverage in the high-rise building is still blank in the academic field and industry. First of all, this paper introduces LTE The system core network and the network architecture of the access network, give the control surface signaling detection through the LTE system interface, and then use the control surface signaling and user surface data to analyze the process of getting the number of users, the traffic volume and the user perception of the Wi Fi information. This paper also studies the principle and characteristics of the time series, and analyzes the analysis of the time series analysis. The principle of ARIMA and the standard of model recognition are widely used, and the prediction steps and evaluation criteria are given. Secondly, this paper analyzes the data characteristics of the traffic volume per hour, and gives a model for predicting the traffic volume based on the analysis. The application conditions of the product season ARIMA model in the prediction of traffic volume are studied, and the model is given. The modeling steps of pattern recognition, order determination and residual check are used to decompose the time series of traffic volume into seasonal, trend and random terms, and then the sequence is seasonally adjusted and then fitted with ETS model, and the prediction results of the latest period are predicted with the data of the latest period. In addition, the Holt-Winters addition is also analyzed. And the multiplication model, and the process of training modeling and prediction using BP neural network. This paper uses these four models to predict the traffic volume and terminal number of a certain operator in a specific scenic area. The experiment shows that the prediction accuracy is good and can meet the requirements of the performance optimization of the mobile network, and it is implemented and applied to the core network of the operators through the code. Finally, after using the LTE network user surface depth packet analysis to obtain the perceived Wi Fi physical address and RSSI in the specific area and building, the Wi Fi information of the high-rise building is obtained through the accumulation of data, and the energy matrix of the Wi Fi is constructed. The element of the matrix is the RSSI value of the Wi Fi collected by the sampling point. Two Wi The correlation coefficient is the correlation between the Wi Fi columns corresponding to the energy matrix, and then the correlation matrix of the Wi Fi is obtained. K-means, PAM, spectral clustering and Fast Unfolding algorithm are used to cluster analysis of Wi Fi samples, and three Wi Fi clusters are obtained. According to the intensity of RSRP measured by the LTE terminal, the coverage quality of mobile users is determined. The experimental results are of high accuracy and can satisfy the optimization requirements of wireless side network.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN929.5

【参考文献】