本文選題：神經網(wǎng)絡 + 復雜網(wǎng)絡　； 參考：《北京郵電大學》2014年碩士論文

【摘要】：近幾年,對下一代網(wǎng)絡的研究已如火如荼,因為下一代網(wǎng)絡所提出的架構理念能很好的解決現(xiàn)在互聯(lián)網(wǎng)所存在的很多問題。與此同時,隨著科技不斷進步,用戶對于所使用的網(wǎng)絡及業(yè)務的服務質量要求不斷提高。所以,立足于用戶需求,再結合下一代網(wǎng)絡的發(fā)展趨勢,我們需要清楚而全面地把握網(wǎng)絡用戶的行為,預測網(wǎng)絡用戶行為的變化規(guī)律,從而為優(yōu)化網(wǎng)絡的性能和提高業(yè)務的服務質量提供一條路徑。 首先,本文介紹了下一代網(wǎng)絡架構；總結了網(wǎng)絡用戶行為的研究現(xiàn)狀；總結了神經網(wǎng)絡用于用戶行為分析及預測的研究現(xiàn)狀。 其次,為了進一步提高網(wǎng)絡流量的預測精度,使模型能自適應不同的業(yè)務流量預測,我們研究了寬參數(shù)域下的回聲狀態(tài)神經網(wǎng)絡算法(ESNs, Echo State Networks)。我們引入復雜網(wǎng)絡理論以及基于生物側抑制機制(LIM,Lateral Inhibition Mechanism)的思想提出了兩種新型的回聲狀態(tài)網(wǎng)絡算法： ·帶有動態(tài)池預測的去耦合回聲狀態(tài)神經網(wǎng)絡(DMESN+RP, Decoupled Mixed Echo State Network with Reservoir Prediction); ·帶有最大信息量的DMESN(DMESN+Maxlnfo,Decoupled Mixed Echo State Network with Maximum Information)。 與此同時,我們與傳統(tǒng)的回聲網(wǎng)絡狀態(tài)算法在預測精度,譜半徑,參數(shù)魯棒性等方面進行了仿真分析及對比。仿真發(fā)現(xiàn)我們所提出的DMESN+RP和DMESN+Maxlnfo在預測精度,譜半徑參數(shù)變化范圍及參數(shù)魯棒性上要優(yōu)于傳統(tǒng)的回聲狀態(tài)神經網(wǎng)絡。 再次,我們將所提出的DMESN+RP和DMESN+Maxlnfo用于移動互聯(lián)網(wǎng)的真實網(wǎng)絡流量預測之中,從預測精度方面驗證這種方案的實用性。 最后,本文結合未來網(wǎng)絡新型分層架構提出了一種基于DMESN+Maxlnfo的網(wǎng)絡節(jié)點流量預測的新型網(wǎng)絡鏈路分配策略。
[Abstract]:In recent years, the research on NGN has been in full swing, because the architecture of NGN can solve many problems existing in the Internet. At the same time, with the development of science and technology, the quality of service of the network and service is improved. Therefore, based on user needs and combined with the development trend of next generation network, we need to clearly and comprehensively grasp the behavior of network users and predict the changing law of network users' behavior. It provides a path for optimizing network performance and improving service quality. Firstly, this paper introduces the next generation network architecture, summarizes the research status of network user behavior, and summarizes the research status of neural network for user behavior analysis and prediction. Secondly, in order to further improve the accuracy of network traffic prediction and enable the model to adapt to different traffic prediction, we study the echo state neural network algorithm ESNs, Echo State networks in wide parameter domain. In this paper, we introduce the theory of complex network and the idea of LIMLlateral Inhibition Mechanism based on the biological side inhibition mechanism. We propose two new echo state network algorithms: De-coupled echo state neural network with dynamic cell prediction (DMESN RP, Decoupled Mixed Echo State Network with Reservoir prediction); DMESN(DMESN Maxlnfol decouped Mixed Echo State Network with Maximum Information with maximum amount of information. At the same time, the simulation analysis and comparison with the traditional echo network state algorithm in prediction accuracy, spectral radius and parameter robustness are carried out. Simulation results show that the proposed DMESN RP and DMESN Maxlnfo are superior to the conventional echo state neural networks in prediction accuracy, spectral radius parameter variation range and parameter robustness. Thirdly, we apply the proposed DMESN RP and DMESN Maxlnfo to the real network traffic prediction of mobile Internet, and verify the practicability of the proposed scheme in terms of prediction accuracy. Finally, this paper proposes a new network link allocation strategy based on DMESN Maxlnfo network node traffic prediction combined with the future network new hierarchical architecture.
【學位授予單位】：北京郵電大學
【學位級別】：碩士
【學位授予年份】：2014
【分類號】：TP393.09

【參考文獻】

相關期刊論文 前10條

1 李擎,鄭德玲,趙星浩,劉東方;一種新的混沌識別方法(Ⅰ)[J];北京科技大學學報;1999年02期

2 張宗震;馬爾可夫預測法基本原理解析與應用[J];成都電子機械高等專科學校學報;2001年02期

3 肖盛;張建華;;基于小世界拓撲模型的電網(wǎng)脆弱性評估[J];電網(wǎng)技術;2010年08期

4 王麗君;劉永強;張健;;基于OpenFlow的未來互聯(lián)網(wǎng)試驗技術研究[J];電信網(wǎng)技術;2011年06期

5 陳敏;葉曉舟;;混沌時間序列的判定方法研究[J];信息技術;2008年06期

6 余少波;嚴雋永;;神經網(wǎng)絡應用[J];海軍工程學院學報;1990年02期

7 范玉妹;玄婧;;ARMA算法在GDP預測中的應用[J];江南大學學報(自然科學版);2010年06期

8 鄒恩,李祥飛,張?zhí)┥?混沌與混沌應用[J];計算機工程與應用;2002年11期

9 裴承丹;;回聲狀態(tài)網(wǎng)絡及其在圖像邊緣檢測中的應用[J];計算機工程與應用;2008年19期

10 李果;吉小恒;;基于小世界拓撲模型的網(wǎng)絡系統(tǒng)安全分析[J];計算機工程;2010年14期

，

本文編號：1971792