大型赛事活动话务预测方法研究

发布时间：2018-08-16 17:03

【摘要】：随着中国的进一步开放和中国经济的增长,越来越多的演唱会、会展、赛事等大型赛事活动开始登陆中国,大量人群的短时聚集对无线网络的建设和配置提出了更高的要求。然而如何设置最合理的硬件配置来达到信道利用率最大化,仍缺乏有效的话务预测方法作为基础,使网络资源配置缺乏依据,也使得大型赛事的保障缺乏理论的突破。大型赛事活动话务预测方法要求训练序列短、输入数据类型少,现有预测方法中时间序列预测法要求下一时刻的话务量与当前时刻和过去时刻的话务量存在相关性,即从长期来看数据序列连续且处于一种稳定的趋势状态,无法直接用于大型赛事活动的话务预测;一般使用的一元线性回归模型仅将预测观众数乘以市场占有率计算得到的预测用户数作为自变量,即使不考虑预测观众人数的准确性,观众并不能完全包含活动区域的全部用户,由于区域长期驻留的用户数并不容易准确统计得到,造成预测精度不稳定。其他预测方法如灰色系统理论、神经网络理论等方法过于复杂,对输入训练序列的长度要求较高,输入参数过多且参数没有明确的理论依据来确定,用于长期话务趋势预测尚可,用于大型赛事活动这类短期突发话务预测难度较大。本文提出了将大型赛事活动期间的话务量拆分成日常部分和活动部分两个相互独立的部分分开进行预测。日常部分具有较多的历史数据可以使用,可以使用如时间序列预测法、神经网络预测法等需要较长训练序列的预测方法;活动部分为离散数据,样本较少,但主要与用户人数有关,可适用回归分析法等预测方法。最后合并两个独立预测部分得到最终的预测结果。这种预测算法简单有效,可用于各类大型赛事活动。
[Abstract]:With the further opening of China and the growth of Chinese economy, more and more large-scale events, such as concerts, exhibitions, events and so on, have begun to land in China. The short time gathering of a large number of people has put forward higher requirements for the construction and configuration of wireless networks. However, how to set up the most reasonable hardware configuration to maximize the channel utilization still lacks the effective traffic prediction method as the foundation, makes the network resource allocation lack the basis, also makes the large-scale competition guarantee lacks the theoretical breakthrough. The traffic prediction methods for large-scale events require short training sequence and few input data types. Among the existing prediction methods, the time series prediction method requires the traffic at the next moment to be correlated with the traffic at the current and past moments. That is, the data sequence is continuous and in a stable trend state in the long run, which can not be directly used for traffic prediction of large-scale events. The commonly used linear regression model only takes the predicted audience number multiplied by the market share as the independent variable, even if the accuracy of the forecast audience size is not considered. The audience can not completely include all the users in the active area. Because the number of users residing in the area for a long time is not easy to get accurate statistics, the prediction accuracy is not stable. Other prediction methods, such as grey system theory, neural network theory and so on, are too complicated to require the length of input training sequence, too many input parameters and no clear theoretical basis to determine the parameters. It is very difficult to predict the trend of long term traffic, and to predict the short term emergency traffic such as large-scale events. In this paper, the traffic during the event is divided into two independent parts, the daily part and the activity part, to be predicted separately. The daily part has more historical data that can be used, such as time series prediction method, neural network prediction method and so on. But mainly related to the number of users, regression analysis and other forecasting methods can be applied. Finally, the final prediction results are obtained by merging the two independent forecasting parts. This prediction algorithm is simple and effective and can be used in all kinds of events.
【学位授予单位】：上海交通大学
【学位级别】：硕士
【学位授予年份】：2011
【分类号】：TP393.09

【参考文献】