類神經網路在寬頻網路管理上之應用

The Applications of Neural Networks on Broadband Network Management

張吉甫(Chi-Fu Chang)

指導教授 ： 李錫捷

元智大學/資訊學院/資訊管理學系/碩士(2002年)

摘要

由於網際網路的便利，帶動網路的蓬勃發展，網路的應用已成為企業主要成功因素之一。然而需求愈殷切，網路也愈龐大和複雜，為有效管理整體網路，就必須有合適的網路管理系統，才能有效執行網路管理(Network Management)的五項功能FCAPS。然而，使用者所關心的主要是網路的訊務(traffic)流量及延遲現象。
本研究之主要目的在建構一有效預測ADSL寬頻網路的訊務流量和頻寬使用率的預測模式，以期作為網路規劃建設及訊務管理之依據。研究之數據來自網管系統及相關資訊系統，使網管系統與類神經網路結合，做到網管不僅提供網路的效能管理，同時也具備訊務預測能力，以增強網管功效。為了克服網路使用訊務量來源的不確定因素，本文以類神經倒傳遞網路(Back-propagation Network)作為網路訊務流量之預測，並將影響輸出流量的輸入變數因素區分為費率類別、傳輸速率、客戶屬性、日期時間四方面。類神經網路具備強大學習能力、非線性、具容錯之優點，預期研究之結果，將可進一步知道使用者特性與訊務流量之關係，以提供決策者作為寬頻網路商場競爭之策略運用。

關鍵字

網路管理 ； 訊務流量預測 ； ADSL ； 類神經網路

並列摘要

In recent years, the convenience of Internet impels the prosperous development of network. The application of network has become one of the key factors that lead the enterprise to success. However, the more the demand is, the more complicated and bulky the network becomes. In order to manage the integrated network efficiently, it is necessary to have a suitable network management system to execute the five FCAPS functions of network management. Nevertheless, what the users mainly concern about is the traffic flow and delay of the network.
The main objective of this study is to build a predictive model to effectively forecast the traffic flow and the bandwidth utilization of ADSL broadband network so that it could be the basis for planning, construction and traffic management of network. The data for research come from network management system and related information systems. Conjoining network management system with neural networks enables network management system not only to provide the efficient management of network, but also owns the ability to predict the traffic load so the power of network management system could be enhanced. To overcome the uncertainty of the traffic load source of network utilization, this study makes use of back-propagation network to predict the traffic flow of network. The input variables that influence the output flow are classified into four types including ISP-class, transmission rate, customer characteristics and date-time. Neural networks have powerful learning capabilities and the advantages of non-linearity and fault-tolerance. The results from the traffic flow forecasting could furthermore realize the relationship between the user characteristics and traffic flow so as to provide the decision-makers as strategy exertion for the competition of broadband network market.

並列關鍵字

Network Management ； Network Traffic Flow Forecasting ； ADSL ； Neural Networks

參考文獻

[3] J. Hall and P. Mars, “ Limitations of artificial neural networks for traffic prediction in broadband networks,” Communication, IEE Proceedings, Volume: 147, Issue 2, pp. 114-118, April 2000.

[4] T. Senjyu et al., “One-Hour-Ahead Load Forecasting Using Neural Network,” IEEE Transaction on Power System, Vol. 17, No. 1, pp. 113-118 February 2002.

[5] S.B. Zhang and Z.M. Liu, “ Neural Network Training Using Ant Algorithm in ATM Traffic Control,” IEEE pp. 157-160, 2001.

[6] E.S. Yu and C.Y.R. Chen, “Traffic prediction using neural networks,” Global Telecommunication Conference, IEEE, Vol. 2, pp. 991-995, 1993.

[7] A.A. Tarraf, I.W. Habib, and S.A. Ahmed, “ATM Multimedia traffic prediction using neural networks,” Global Data Networking, IEEE Proceedings, pp. 77-84, 1993.