於LTE-A網路中基於機器學習之自我錯誤診斷及修復

Abstract

隨著商用無線行動網路的複雜度及服務的使用者數增加，電信供應商及服務商如何維護基地台和維持一定的服務品質將成為一大挑戰。因此，3GPP在LTE標準中引入自我組織網路的概念。自我組織網路即為基地台根據收到使用者的一些參數去做自我配置、自我優化及自我修復，以達到減少資本支出及營運費用。於本論文中，吾人提出在LTE-A中使用監督式機器學習及類神經網路演算法去做單一問題判別，此方法利用基地台收集到個別不同的網路狀況所產生的性能指標參數經過吾人設計的通用預處理演算法後，饋入機器學習及類神經演算法訓練，使得基地台擁有單一問題判別能力。此外，考慮多問題同時發生的情況，吾人提出結合吾人設計的預處理演算法、多標籤機器學習演算法及類神經演算法達到多問題同時發生及問題嚴重程度的診斷。最後，由分析和電腦模擬結果顯示，吾人提出的演算法能夠提高判斷的精準度，以達到基地台自我修復的功能。

Recently, mobile networks have been designed with more complexity of infrastructure and higher diversity of associated devices and resources, as well as dynamical formations of networks due to demand of user and development of the mobile industry. Therefore, the infrastructure of the mobile system is hard to be maintained, and cost of maintenance is going to increase. In many works, algorithms for automatic configuration, optimization and fault healing have been proposed to solve these problems mentioned above. In this thesis, self-healing is focused. First, we propose a method that uses supervised machine learning and neural network algorithms to discriminate single-labeled and level of severity for anomalies. Afterward, a method for identifying single-labeled and multi-labeled anomalies is proposed. These methods utilize key performance indicators (KPIs) of different network conditions, which base stations collect and use for preprocessing. Subsequently, the data, which is passed through preprocessing, is used as an information source for machine learning through the neural network, given that base stations have the ability to discriminate single-labeled and multi-labeled anomalies. In the final analysis, computer simulation results show that our proposed algorithms can enhance the accuracy of anomaly discrimination such that base stations are able to self-heal.