【作者】 徐伟强；

【导师】 王敏锡；

【作者基本信息】 西南交通大学 ， 通信与信息系统， 2002， 硕士

【摘要】 随着移动电话在中国的普及，对话务量需求的预测日益受到移动运营商的关注。预测结果的准确与否直接关系到企业未来的发展，因此对话务量需求预测的研究变得日益迫切。本文在该领域主要作了以下几点研究。 1．比较系统地对国际上已经研究的各种预测方法进行了归纳和综述，为本文研究提供了依据。 2．话务量需求是受多种因素影响的非线性系统。本文运用相空间重构技术求取了最佳嵌入延迟和嵌入维的估计值，运用Wolf算法估算出最大Lyapunov指数和最大可预测时间。研究结果表明了话务量时间序列具有混沌特性及可预测性。 3．由于移动话务量的时间序列表现出复杂的非线性混沌特性，而神经网络具有很强的非线性映射能力，理论上能以任意精度逼近任意形式的非线性函数，而且具有很强的学习能力，因此采用了神经网络对话务量时间序列进行预测。先运用BP网络对其进行预测。之后，考虑到话务量时间序列既具有混沌特性，同时也具有随机、趋势、周期循环等特性，为提高预测效果，应用了动态特性较强的递归神经网络—Elman网络，提高了预测精度。 4．任何单一的预测法所获的信息是有限的，因此该类方法要冒较大的风险。为此，提出了将几种不同预测方法取得的预测值加权平均后作为最终预测值的组合预测法。考虑到遗传算法具有全局优化、并行处理、通用性好、稳定性强等优点，本文运用参数自适应实数遗传算法来求得组合预测的权系数，仿真的结果表明了预测精度有所提高。 5．比较系统地提出了移动通信网话务量需求预测系统的设计思路。 上述几方面的研究，使移动运营商预测话务量需求有了理论指导，也为国内移动通信网话务量需求预测的研究作了有益的探索。更多还原

【Abstract】 With the popularity of mobile telephone in China, mobile operator has been paying more and more attention to the prediction of the telephone traffic demand. The accuracy of the prediction result directly concerns the future development of enterprise. Therefore, research on the prediction of the telephone traffic demand is increasingly urgent. In this paper, the following aspects have been mainly studied.1. The various international prediction methods which have offered basis for the research in this paper are summarized.2. The telephone traffic demand is nonlinear system affected by various factors. The value of the optimal delay time lag and embedding dimension are estimated through phase space reconstruction. Largest Lyapunov exponent and predictable time are acquired by applying Wolf algorithm. The research result has shown Chaos property and predictability of the time-series of the telephone traffic demand.3. Since the time-series of the telephone traffic demand show complex nonlinearity and chaos property, and neural network has the strong ability to map nonlinear and can theoretically approach nonlinear functions in any form with high precision, and has the powerful learning ability, therefore, neural network is applied to the prediction of the time-series of the telephone traffic demand. Firstly, Back-Propagation neural network is used. Then, recurrent neural network ?Elman network which has more dynamic property is adopted to predict the telephone traffic demand, because the time-series of the telephone traffic demand not only has chaos property , but also has the properties such as tendency and periodicity. It has raised prediction precision.4. Any single prediction method is confronted with the enormous risk of prediction error owing to the limitation of information used. Combined prediction method, whose prediction value is acquired through putting different weight to theprediction value of some kinds of prediction methods, therefore, has been put forward. Because genetic algorithm has advantages such as the optimization of overall situation, parallel handling property and versatility, strong stability etc, the adaptive parameter real-coded genetic algorithm(APRGA) is introduced to solve the weight coefficients of combined prediction. The result of simulation has shown that the precision of prediction is raised.5. Design idea of prediction system of telephone traffic demand in mobile communication networks has also been systematically put forward.What has been mentioned above is not only theoretical guidance to predict the telephone traffic demand for mobile operator, but also beneficial beginning of the research on prediction of telephone demand traffic in mobile communication networks.更多还原