Dynamic IoT management system using K-means machine learning for precision agriculture applications

Abstract

Multi-media applications for use in Precision Agriculture (PA) and Smart Farming (SM) require Network Management Systems to deliver Quality of Service (QoS) end-to-end guarantees. This paper presents the second phase of the research in providing a network management system capable of delivering end-to-end QoS guarantees for Internet of Things (IOT) networks. The first phase of this work used a wireless test bed to develop a propagation model to incorporate the attenuation due to foliage in dense vegetation typically found in PA environments. The output of this propagation model will influence the decision making process in the network management system. Wireless Multimedia Sensor Networks (WMSN) operate under the umbrella of the Wireless Sensor Network (WSN) IEEE 802.15.4 Medium Access Control (MAC) and Physical (PHY) protocol to deliver multimedia applications such as voice, video and live streaming. To operate successfully these multi-media applications have high QoS requirements. To enable these QoS requirements to be fulfilled performance metrics such as throughput, end-to-end delay and limited packet loss must be guaranteed. This next phase of the work in developing the intelligent network management system presented in this paper uses an OPNET™ simulation package to implement a modified K-Means algorithm to detect the presence of multi-media traffic. Consequently a signal informs the network management system to adopt pre-configured settings via the Personal Area Network Co-ordinator (PANC). The resulting changes implement service differentiation by manipulating the MAC layer (size of the individual GTS timeslots and duty cycle) to deliver better throughput and end-to-end delay performance. OPNET™ simulation results show that the new algorithm facilitates better performance and meets QoS requirements suitable to multimedia applications. This paper focuses on the derivation and evaluation of the performance of the K-Means algorithm. The sensory nodes are power, memory and computationally restricted. These restrictions coupled with the heterogeneous structure of the wireless network make intelligent network management systems very important if the QoS requirements are to be fulfilled. Upon detection of multimedia traffic with high QoS demands usually triggered in the aftermath of an event of particular interest e.g. security threat etc., a management system must dynamically effect change of the network configuration settings to maintain such guarantees. As real-time applications require an urgent response, the dynamic change must occur during run time automatically. This research work is novel in that it combines the output from the development of the physical layer propagation model to inform a network management system to trigger service differentiation for multimedia traffic in a PA environment.