基於機器學習之物聯網即時性分類技術

Abstract

隨著機器類型通訊（Machine Type Communication）和長期演進技術（LTE）和先進長期演進技術（LTE-A）這樣的先進胞狀網路（Cellular Network）的普及，LTE已經被驗證是機器類型通訊的良好通訊方式。一種加強的LTE MTC閘道器通訊架構被提出。大量的研究表明，在IoT裝置數量日益增長的今天，上行封包壅塞的問題越來越嚴重。儘管在許多研究在無線電資源分配（Radio Resource Allocation）技術上取得了不錯的結果，但是這需要通訊協定的更改，並將會花費大量的社會資源。這篇論文討論物聯網裝置在發生封包上行壅塞的情況下，只通過更改閘道器的架構就能夠得到更好的效能並滿足即時性需求。在這篇論文中，我們設計了結合監督式學習和非監督式學習技術的優先度標籤系統。這個系統可以確定物聯網裝置是否處於有即時性需求的狀態並標記它們，我們將它稱之為IoT real-time traffic classifier (IRTC)。實驗表明，通過IRTC我們可以在平均8.63到13.84個封包延遲後偵測到物聯網裝置的狀態變化。

As the development of Machine type communications (MTC) and advanced cellular network, such as long-term evolution (LTE) and LTE-Advanced (LTE-A), LTE has been proved as a suitable communication protocol for MTC. An enhanced LTE MTC gateway communication architecture has been proposed. Researches show that with the numbers of IoT devices increasing, the uplink congestion problem becomes severe. Although amounts of radio resource allocation schemes have been proposed in order to provide a high-performance effective MTC, radio resource allocation requires protocol modification, which costs social resource. To our knowledge, no paper considers using IoT devices priority and real-time requirement to solve the uplink congestion problem in LTE MTC gateway. In this thesis, we designed a classification system by combing supervised learning and unsupervised learning model. This system named IRTC (IoT real-time traffic classifier) can determine whether IoT devices are in the state with real-time requirement state and then labels them. Experimental result shows that we can determine the state transition within 8.63 to 13.84 packets delay by IRTC