【作者】 王喆；

【导师】 张丽翠；

【作者基本信息】 吉林大学 ， 电子与通信工程（专业学位）， 2015， 硕士

【摘要】 通信技术的发展已经使得无线网络通信的承载技术不再是制约无线网络发展的关键因素。另一方面，各大网络运营商为提高用户对网络性能的满意程度，正在采用各种网络性能指标检测方法和各种故障预测和排查办法。而针对网络性能指标数据的检测，目前尚停留在只检测或部分用于分析网络运维态势的状况。也就是说，监测到的网络性能数据没有得到充分地利用，用以提高对网络运行状态的掌控力度和预判准确度。针对TD-SCDMA中网络性能指标数据的特点，针对一些性能指标存在理论上的相关性的问题，TD-SCDMA网络的运行状态管理的需求，本文对网络性能的分析方法和预测方法进行了研究。主要研究了人工神经网络分析预测方法和基于多元时间序列的分析预测方法。网络性能分析是指根据网络运行中采集到的历史数据分析网络的运行状况，并对网络运行的未来趋势做出相应预测。网络性能预测是网络优化，网络故障预处理的基础。在网络性能指标数据理论的研究基础上，详细学习了误差反向传播（BP）神经网络预测算法的基本原理和预测算法流程，实现网络性能指标话务量的预测分析。在研究了神经网络的学习规则和预测原理的基础上，我们主要研究误差反向传播神经网络预测算法流程，学习规则及其误差传播和修正过程，并通过仿真实验对TD-SCDMA的网络性能指标话务量进行了分析和预测。实验结果表明，误差反向传播神经网络的预测精度在90%以上，并且预测精度与输入向量的维数成正相关关系，算法执行速度能够满足话务量性能指标数据的实时预测要求。由于网络性能指标数据可以看作为时间序列数据，因此，我们采用时间序列预测方法来实现网络性能指标接通率的预测。本文在模糊时间序列预测算法部分详细的描述了模糊时间序列的符号化方法、相关性计算方法、模糊关系矩阵建立方法，模式匹配规则。在仿真实验中，通过平稳性检验判断话务量性能指标数据经过归一化，取对数，差分等变换后仍然不满足平稳性条件，而接通率和拥塞率性能指标经过一阶差分后均为平稳序列。因此，在仿真试验中，对一阶单整的接通率和拥塞率性能指标数据进行了二元模糊时间序列的预测，实现了接通率性能指标数据的预测。通过二元模糊时间序列的预测实例，验证了本文所提出的模糊时间序列算法的预测有效性。更多还原

【Abstract】 With the development of communication technology, the bearing technology ofthe wireless network communication is no longer the key factor that restricts thewireless network development. On the other hand, in order to improve the usersatisfaction with the performance of the network, each network operating comprise isundertaking various network performance testing methods and all kinds of faultprediction and screening methods. While, specific to the network performanceindicators, the discourage stage is that we only monitor the network performance dataand only a small part is applied to the analysis of the status of the network operationalsituation. That is to say, the detected network performance data has not been fullyutilized, such as to improve its control of the network running status and theforecasting accuracy. Specific for the features of network performance indicator datain TD-SCDMA, as well as the theoretically relevance of performance indicators, andto meet the management need of TD-SCDMA network, the network performanceanalysis and prediction method are researched, mainly including the artificial neuralnetwork based analysis method and the multivariate time series prediction method.Network performance analysis is to recognize the network running status basedon the historical data, which is collected in the network operation, and to forecast thefuture trend of the network operation accordingly. Network performance prediction isthe foundation of network optimization and network fault pretreatment. On the basisof studying the theory of network performance indicator data, we detailed study thebasic principle and the prediction algorithm process of error back propagation (BP)neural network prediction algorithm, and we use it to realize the phone trafficforecasting analysis. After studying the learning rule of neural network and predictionprinciple, we mainly research the algorithm process of error back propagation network prediction, and its learning rules and error propagation and correction process.Simulation experiment analyzed and forecast the TD-SCDMA network trafficperformance indicators. Experimental results show that the error back propagationneural network prediction accuracy is above90%, and the prediction precision ispositively correlated with the dimensions of the input vector. And the algorithmexecution speed can meet the requirement of real-time phone traffic performanceindicator data prediction.The network indicators can be seen as time series, so we use time seriesforecasting method to realize the prediction of network performance indicator. In thesection of fuzzy time series prediction algorithm, we describe the symbolization offuzzy time series method, correlation calculation method, the fuzzy rule matrixconstruction and the matching rules. In the simulation, we judge the smoothness ofthe phone traffic performance data after normalizing, log calculating, differencing.Connection rate and congestion rate are stationary series after first order differencing.Then in the simulation test, a binary fuzzy time series prediction based on connectionrate and congestion rate is conducted to realize the connection rate prediction.Through the example of binary fuzzy time series prediction, it verified theeffectiveness of the proposed fuzzy time series prediction algorithm.更多还原