TD-SCDMA网络性能预测及监控系统的研究与实现
|
摘要
TD-SCDMA网络性能预测分析及优化系统是面向3G及以上网络的综合性能分析和优化平台，本项目将弥补传统通信网络管理系统的不足。采用高端的软件工程技术和先进的管理思想，实现TD-SCDMA交换网络实时性能预测分析及主动监控系统，根据业务情况主动监控网络指标数据，对业务管理提出建议，为运维管理面向业务发展找到真正的解决方案。
本文工作主要完成TD-SCDMA网络性能预测分析和监控系统的研究与实现。网络性能预测是主动监控系统的一个重要组成部分，它既可用于对未来一天或几天的短期预测，也能进行一月或一年的长期预测，以此了解性能指标的发展趋势。预测对于网络管理人员提前解决即将发生的故障或发现未来存在的隐患，在交换机告警发生前及时的采取措施，防范于未然，及时优化网络。本文在性能分析中采用了一种新型的预测方法：基于中值滤波的高斯过程回归模型应用于网络性能指标的预测。
主动监控系统是TD-SCDMA网络性能预测分析及优化系统的重要组成部分，是预测分析和指标监控告警的主体。其中“主动监控实时性能告警算法”是该系统监控告警模块的核心，由“基线算法”、“容忍度计算方法”、“告警机制”三部分共同组成。
Along with the rapid development of mobile communication networks and the increasingcompetition in the market, the pros and cons of network quality have become a key factor toimprove the competitiveness of enterprises. Therefore, greater attentions have paid to themeasurement of build level of itself network quality and customer satisfaction by the mobileoperators. The rapid development of mobile communications business will pose greaterchallenges to the network. The rapid growth of network users and applications and theoverloading of network equipment, promote increasingly heavy burden of the network. Thelack of reliable technologies and means on network planning and network optimization,results in decreased network performance. Therefore, the evaluation system is needed to dealwith these problems, which not only predict and analyze network performance, but alsoreal-time monitor it as much as possible to avoid problems, ameliorate network performanceand improve operational quality, then satisfied the customer.
The existing communications operator's network management doesn’t technically realize"active monitoring" of communication network according to user needs and businessdevelopment, and can not predict the impending failure of the network, and take preventivemeasures before the occurrence of the switch alarms to optimize network in a timely manner.In addition, business and operation and maintenance management system are independent ofeach other, with no automatic process flow and data sharing. So it can not achieve theproactive monitoring of network management for business to format the network dynamicalarm threshold in real time based on business.
TD-SCDMA network performance prediction analysis and optimization system is aplatform for3G and above that it can assist network managers implement network qualityassessment, network planning and performance optimization quickly and efficiently, sinceextensive and in-depth intelligent analysis be conducted based on large amount of historicaldata such as the parameters of configuration information, a variety of performance indicatorsand alarms, and a wealth of performance prediction, statistical computing, network planning,graphical display of report output and problems seeking functions be performed. The projectwill compensate for the deficiencies of the traditional communication NMS. High-endsoftware engineering technology and advanced management thinking, implementingTD-SCDMA switched network real-time performance analysis and active monitoring system,proactively monitoring the network performance according business, makingrecommendations on business management, will make a real solution for operation andmaintenance management oriented business development.
Network performance prediction is an important part of active surveillance system thatoptimize the network in a timely manner to predict the approximate trend of the performanceindicators for the next hours, day or few days, for the impending failure be solved in advanceby the systems and even network management staff. Measures are prevented by take measuresbefore the switch alarm timely.
It is always difficult to accurately predict System for Mobile Communications networkperformance, for it is affected by a number of factors. A variety of methods and techniques,such as the gray prediction, BP neural network, time series analysis, regression analysis, alsowidely used in network performance prediction, and each method has its own advantages anddisadvantages. In this paper, a new prediction method, Gaussian process regression modelbased on the median filtering, is applied to the prediction of network performance indicators.Compared with neural network prediction method, Gaussian process have fewer modelparameters, easier to parameter optimization problem, and to convergence, predict effect rapidand apparently. However, it brings a number of unexpected challenges to the Gaussianprocess regression model prediction to make the effect of prediction undesirable after trainingwith large deviation of predictions, due to several factors such as fluctuations in networkperformance indicators, stochastic and nonlinear strongly. In this paper, by using medianfiltering, fluctuations indicator data become relatively smooth after filtered, trends is morepronounced. This will give a significant improvement for the training of the GP modellearning, and make the predictive value more accurate. Predictive analysis features havesubstantial, effective, and rational function and role for network managers or optimize staff.Experiments show that the combination method, presented by this paper, improves theprediction accuracy with a strong self-learning ability, and provides a new method and meanfor network performance analysis.
Active monitoring system is an important part of the network performance predictionanalysis and optimize system of TD-SCDMA, is the main body of the prediction analysis andindicators monitoring alarm, of which the "active monitoring real-time performance alarmalgorithm" is core of the system monitoring alarm module, consisting of the "baselinealgorithm","tolerance calculation method" and "alarm mechanism".The algorithm comes tureby comparing the collected performance indicators data with the predefined tolerate line(alarm trigger threshold),when it is bigger than the tolerate line,the alarm generationmechanism be triggered.
At last in this paper the completed work have been summarized simplely.And it givesviews and outlook for the further experiment research of prediction.
本文工作主要完成TD-SCDMA网络性能预测分析和监控系统的研究与实现。网络性能预测是主动监控系统的一个重要组成部分，它既可用于对未来一天或几天的短期预测，也能进行一月或一年的长期预测，以此了解性能指标的发展趋势。预测对于网络管理人员提前解决即将发生的故障或发现未来存在的隐患，在交换机告警发生前及时的采取措施，防范于未然，及时优化网络。本文在性能分析中采用了一种新型的预测方法：基于中值滤波的高斯过程回归模型应用于网络性能指标的预测。
主动监控系统是TD-SCDMA网络性能预测分析及优化系统的重要组成部分，是预测分析和指标监控告警的主体。其中“主动监控实时性能告警算法”是该系统监控告警模块的核心，由“基线算法”、“容忍度计算方法”、“告警机制”三部分共同组成。
Along with the rapid development of mobile communication networks and the increasingcompetition in the market, the pros and cons of network quality have become a key factor toimprove the competitiveness of enterprises. Therefore, greater attentions have paid to themeasurement of build level of itself network quality and customer satisfaction by the mobileoperators. The rapid development of mobile communications business will pose greaterchallenges to the network. The rapid growth of network users and applications and theoverloading of network equipment, promote increasingly heavy burden of the network. Thelack of reliable technologies and means on network planning and network optimization,results in decreased network performance. Therefore, the evaluation system is needed to dealwith these problems, which not only predict and analyze network performance, but alsoreal-time monitor it as much as possible to avoid problems, ameliorate network performanceand improve operational quality, then satisfied the customer.
The existing communications operator's network management doesn’t technically realize"active monitoring" of communication network according to user needs and businessdevelopment, and can not predict the impending failure of the network, and take preventivemeasures before the occurrence of the switch alarms to optimize network in a timely manner.In addition, business and operation and maintenance management system are independent ofeach other, with no automatic process flow and data sharing. So it can not achieve theproactive monitoring of network management for business to format the network dynamicalarm threshold in real time based on business.
TD-SCDMA network performance prediction analysis and optimization system is aplatform for3G and above that it can assist network managers implement network qualityassessment, network planning and performance optimization quickly and efficiently, sinceextensive and in-depth intelligent analysis be conducted based on large amount of historicaldata such as the parameters of configuration information, a variety of performance indicatorsand alarms, and a wealth of performance prediction, statistical computing, network planning,graphical display of report output and problems seeking functions be performed. The projectwill compensate for the deficiencies of the traditional communication NMS. High-endsoftware engineering technology and advanced management thinking, implementingTD-SCDMA switched network real-time performance analysis and active monitoring system,proactively monitoring the network performance according business, makingrecommendations on business management, will make a real solution for operation andmaintenance management oriented business development.
Network performance prediction is an important part of active surveillance system thatoptimize the network in a timely manner to predict the approximate trend of the performanceindicators for the next hours, day or few days, for the impending failure be solved in advanceby the systems and even network management staff. Measures are prevented by take measuresbefore the switch alarm timely.
It is always difficult to accurately predict System for Mobile Communications networkperformance, for it is affected by a number of factors. A variety of methods and techniques,such as the gray prediction, BP neural network, time series analysis, regression analysis, alsowidely used in network performance prediction, and each method has its own advantages anddisadvantages. In this paper, a new prediction method, Gaussian process regression modelbased on the median filtering, is applied to the prediction of network performance indicators.Compared with neural network prediction method, Gaussian process have fewer modelparameters, easier to parameter optimization problem, and to convergence, predict effect rapidand apparently. However, it brings a number of unexpected challenges to the Gaussianprocess regression model prediction to make the effect of prediction undesirable after trainingwith large deviation of predictions, due to several factors such as fluctuations in networkperformance indicators, stochastic and nonlinear strongly. In this paper, by using medianfiltering, fluctuations indicator data become relatively smooth after filtered, trends is morepronounced. This will give a significant improvement for the training of the GP modellearning, and make the predictive value more accurate. Predictive analysis features havesubstantial, effective, and rational function and role for network managers or optimize staff.Experiments show that the combination method, presented by this paper, improves theprediction accuracy with a strong self-learning ability, and provides a new method and meanfor network performance analysis.
Active monitoring system is an important part of the network performance predictionanalysis and optimize system of TD-SCDMA, is the main body of the prediction analysis andindicators monitoring alarm, of which the "active monitoring real-time performance alarmalgorithm" is core of the system monitoring alarm module, consisting of the "baselinealgorithm","tolerance calculation method" and "alarm mechanism".The algorithm comes tureby comparing the collected performance indicators data with the predefined tolerate line(alarm trigger threshold),when it is bigger than the tolerate line,the alarm generationmechanism be triggered.
At last in this paper the completed work have been summarized simplely.And it givesviews and outlook for the further experiment research of prediction.
引文
[1]陈艳敏.电子信息产业技术创新实力稳步提高[N].中国电子报,2010-10-19(001).
[2]黄琅,周超.通信网络性能分析系统的设计与实现[J].中国水运（理论版）,2006,4(4):124-125.
[3]李旭.移动通信网络性能分析系统的设计与实现[D].北京邮电大学硕士学位论文,2009,1.
[4]曾彬.基于主动测试的网络性能监测技术研究[D].湖南大学硕士学位论文,2009,1
[5]涂光友.数据挖掘在网络故障诊断中的应用研究[D].重庆大学硕士学位论文,2004,1.
[6]陈曦,高海东.移动通信中的常用业务预测方法[C].中国通信学会第五届学术年会论文集,2008,671-675.
[7]刘童.话务量时间序列预测方法的实现[D].吉林大学硕士学位论文,2008.
[8]李强,马东堂.混合预测模型在网络性能预测中的应用[J].计算机工程,2009,35(13):88-90.
[9]陶伟宜.基于回归方法的移动通信话务量预测[J].邮电设计技术,2004,5:29-32.
[10]马曼.神经网络在话务量预测中得应用研究[D].吉林大学硕士学位论文,2009.
[11]徐伟强.移动通信网话务量需求的混沌特性及预测方法研究[D].西南交通大学研究生学位论文,2000.
[12]杨玉美.组合预测方法在山东省电信业务收入中得实践研究[D].辽宁科技大学硕士学位论文,2010.
[13]邓德.基于禁忌算法的长沙移动网络规划软件[D].湖南大学硕士学位论文,2009,4.
[14]Jack M.Wang, David J. Fleet, Aaron Hertzmann. Gaussian Process Dynamical Models[J].University of Toronto,2005,1-8.
[15]Carl Edward Rasmussen. Gaussian Processes in Machine Learning[J]. Max PlanckInstitute for Biological Cybernetics,72076T¨ubingen, Germany,2008,67-75.
[16]曾斌,杨屹等.高斯回归过程学习方法在TC4流变应力预测中的应用[J].塑性工程学报,2010,17(1):110-114.
[17]申倩倩，孙宗海.基于自适应自然梯度法的在线高斯过程建模[J].计算机应用研究,2011,28(1):95-97.
[18]Zhang Wei, Wu Chunf eng, Deng Pan, Fan Ning. Research on Gaussian distributionpreprocess method of infrared multispectral image background clutter. J ournal of H arbin Inst itute of Technology (New Series),2004,11(5):513-515.
[19]Carl Edward Rasmussen,Christopher K. I. Williams. Gaussian Processes for MachineLearning[M]. Massachusetts:MITPress，2006.
[20]Lu C,Wang Y.The Local Continuity Moduli For Two Classes Of Gaussian Processes[J].App l. Math. J. Chinese Uni v. Ser. B,2000,15(2):161-166.
[21]Guohao Su.Modeling Non-linear Deformation Time Series Of Tunnel Using GaussianProcess Machine Learning[C].ISRM Inernational Symposium on RockMechanics,2009,150.
[22]苏国韶,燕柳斌.基坑位移时间序列预测的高斯过程方法[J].广西大学学报(自然科学版),2007,32(2):223-226.
[23]Guoshao Su.Accelerating Particle Swarm Optimization Algorithms Using GaussianProcess[J].2009International Conference on Computational Intelligence and NaturalComputingMachine Learning,2009,174-177.
[24]周占杰.伽玛函数和贝塔函数在概率统计中的应用[J].电大理工,2009,238(1):59-62.
[25]苏国韵.圆弧破坏型岩质边坡安全系数快速估计的高斯过程模型[J].应用基础与工程科学学报,2010,18(6):959-966.
[26]单志辉.基于高斯过程回归的翼型快速设计研究[D].南京航空航天大学硕士学位论文,2011,9-14.
[27]Michael Osborne. Gaussian Processes for Prediction[D]. Robotics Research GroupDepartment of Engineering Science University of Oxford,2007,13-32.
[28]M. Todd Farrell,Andrew Correa. Gaussian Process Regression Models for PredictingStock Trends.2007:2-3.
[29]王红雷,何成源.行半正定矩阵的充要条件及其性质[J].数学理论与应用,2011,31(2):12-14.
[30]Yan Zhang, Guoshao Su and Liubin Yan. Gaussian Process Machine Learning Model forForecasting of Karstic Collapse[C].20103rd International Conference on ComputationalIntelligence and Industrial Application (PACIIA),2010,209-212.
[31]Nicos A,James C. Bayesian Learning of Bayesian Networks with Informative Priors[M].Nicos A,James C. Bayesian Learning of Bayesian Networks with InformativePriors,2010,1-5.
[32]Zhou Yatong, Zhang Taiyi,Li Xiaohe. MULTI-SCALE GAUSSIAN PROCESSESMODEL[J].JOURNAL OF ELECTRONICS (CHINA)，2006,23(4):618-622.
[33]李渝燕.基于第三代移动通信系统的无线网络性能评估及优化[D].山东大学硕士学位论文,2006,12-13.
[34]GUO Qingchun,HE Zhenfang,LI Li. A Prediction Model of Peasants’Income in ChinaBased on BP Neural Network[J]. Asian Agricultural Research.2011,3(4):88-90,94.
[35]王晓敏,刘希玉,戴芬.BP神经网络预测算法的改进及应用[J].计算机技术与发展,2009,19(11):64-67.
[36]刘国宏,郭文明.改进的中值滤波去噪算法应用分析[J].计算机工程与应用,2010,46(10):187-189.
[37]路鹏,刘财等.中值滤波技术在遥感图像处理中的应用[J].吉林大学学报(地球科学版),2005,35:151-154.
[38]Liu J,Li S.Li zy,Caixia lv.Model Analysis of Image Median Filter Based on NoC[C].20102rd International Conference on Power Electronics and Intelligent TransportationSystem,2010,336-339.
[39]靳莉.基于WEB的网络性能监控系统的研究与实现[C].中国科技论文在线.
[40]周乔奇.面向客户感知的移动通信网络质量评估体系的研究[D].复旦大学硕士学位论文,2009,48-51.
[2]黄琅,周超.通信网络性能分析系统的设计与实现[J].中国水运（理论版）,2006,4(4):124-125.
[3]李旭.移动通信网络性能分析系统的设计与实现[D].北京邮电大学硕士学位论文,2009,1.
[4]曾彬.基于主动测试的网络性能监测技术研究[D].湖南大学硕士学位论文,2009,1
[5]涂光友.数据挖掘在网络故障诊断中的应用研究[D].重庆大学硕士学位论文,2004,1.
[6]陈曦,高海东.移动通信中的常用业务预测方法[C].中国通信学会第五届学术年会论文集,2008,671-675.
[7]刘童.话务量时间序列预测方法的实现[D].吉林大学硕士学位论文,2008.
[8]李强,马东堂.混合预测模型在网络性能预测中的应用[J].计算机工程,2009,35(13):88-90.
[9]陶伟宜.基于回归方法的移动通信话务量预测[J].邮电设计技术,2004,5:29-32.
[10]马曼.神经网络在话务量预测中得应用研究[D].吉林大学硕士学位论文,2009.
[11]徐伟强.移动通信网话务量需求的混沌特性及预测方法研究[D].西南交通大学研究生学位论文,2000.
[12]杨玉美.组合预测方法在山东省电信业务收入中得实践研究[D].辽宁科技大学硕士学位论文,2010.
[13]邓德.基于禁忌算法的长沙移动网络规划软件[D].湖南大学硕士学位论文,2009,4.
[14]Jack M.Wang, David J. Fleet, Aaron Hertzmann. Gaussian Process Dynamical Models[J].University of Toronto,2005,1-8.
[15]Carl Edward Rasmussen. Gaussian Processes in Machine Learning[J]. Max PlanckInstitute for Biological Cybernetics,72076T¨ubingen, Germany,2008,67-75.
[16]曾斌,杨屹等.高斯回归过程学习方法在TC4流变应力预测中的应用[J].塑性工程学报,2010,17(1):110-114.
[17]申倩倩，孙宗海.基于自适应自然梯度法的在线高斯过程建模[J].计算机应用研究,2011,28(1):95-97.
[18]Zhang Wei, Wu Chunf eng, Deng Pan, Fan Ning. Research on Gaussian distributionpreprocess method of infrared multispectral image background clutter. J ournal of H arbin Inst itute of Technology (New Series),2004,11(5):513-515.
[19]Carl Edward Rasmussen,Christopher K. I. Williams. Gaussian Processes for MachineLearning[M]. Massachusetts:MITPress，2006.
[20]Lu C,Wang Y.The Local Continuity Moduli For Two Classes Of Gaussian Processes[J].App l. Math. J. Chinese Uni v. Ser. B,2000,15(2):161-166.
[21]Guohao Su.Modeling Non-linear Deformation Time Series Of Tunnel Using GaussianProcess Machine Learning[C].ISRM Inernational Symposium on RockMechanics,2009,150.
[22]苏国韶,燕柳斌.基坑位移时间序列预测的高斯过程方法[J].广西大学学报(自然科学版),2007,32(2):223-226.
[23]Guoshao Su.Accelerating Particle Swarm Optimization Algorithms Using GaussianProcess[J].2009International Conference on Computational Intelligence and NaturalComputingMachine Learning,2009,174-177.
[24]周占杰.伽玛函数和贝塔函数在概率统计中的应用[J].电大理工,2009,238(1):59-62.
[25]苏国韵.圆弧破坏型岩质边坡安全系数快速估计的高斯过程模型[J].应用基础与工程科学学报,2010,18(6):959-966.
[26]单志辉.基于高斯过程回归的翼型快速设计研究[D].南京航空航天大学硕士学位论文,2011,9-14.
[27]Michael Osborne. Gaussian Processes for Prediction[D]. Robotics Research GroupDepartment of Engineering Science University of Oxford,2007,13-32.
[28]M. Todd Farrell,Andrew Correa. Gaussian Process Regression Models for PredictingStock Trends.2007:2-3.
[29]王红雷,何成源.行半正定矩阵的充要条件及其性质[J].数学理论与应用,2011,31(2):12-14.
[30]Yan Zhang, Guoshao Su and Liubin Yan. Gaussian Process Machine Learning Model forForecasting of Karstic Collapse[C].20103rd International Conference on ComputationalIntelligence and Industrial Application (PACIIA),2010,209-212.
[31]Nicos A,James C. Bayesian Learning of Bayesian Networks with Informative Priors[M].Nicos A,James C. Bayesian Learning of Bayesian Networks with InformativePriors,2010,1-5.
[32]Zhou Yatong, Zhang Taiyi,Li Xiaohe. MULTI-SCALE GAUSSIAN PROCESSESMODEL[J].JOURNAL OF ELECTRONICS (CHINA)，2006,23(4):618-622.
[33]李渝燕.基于第三代移动通信系统的无线网络性能评估及优化[D].山东大学硕士学位论文,2006,12-13.
[34]GUO Qingchun,HE Zhenfang,LI Li. A Prediction Model of Peasants’Income in ChinaBased on BP Neural Network[J]. Asian Agricultural Research.2011,3(4):88-90,94.
[35]王晓敏,刘希玉,戴芬.BP神经网络预测算法的改进及应用[J].计算机技术与发展,2009,19(11):64-67.
[36]刘国宏,郭文明.改进的中值滤波去噪算法应用分析[J].计算机工程与应用,2010,46(10):187-189.
[37]路鹏,刘财等.中值滤波技术在遥感图像处理中的应用[J].吉林大学学报(地球科学版),2005,35:151-154.
[38]Liu J,Li S.Li zy,Caixia lv.Model Analysis of Image Median Filter Based on NoC[C].20102rd International Conference on Power Electronics and Intelligent TransportationSystem,2010,336-339.
[39]靳莉.基于WEB的网络性能监控系统的研究与实现[C].中国科技论文在线.
[40]周乔奇.面向客户感知的移动通信网络质量评估体系的研究[D].复旦大学硕士学位论文,2009,48-51.