DEEP LEARNING APPROACH FOR EFFICIENT ENERGY CONSUMPTION AND HIGH THROUGHPUT IN MOBILE WIRELESS SENSOR NETWORKS

Abstract

Features such as scalability, smaller size, simplicity, low-cost operation, selforganization abilities, and easy and fast deployment are the main parameters of a Wireless Sensor Network (WSN). The research demand is growing on WSNs, and therefore, areas under agriculture, industry, healthcare, manufacturing, security, surveillance, transport, air quality, water quality, etc., have started to possess the attributes of WSNs. The primary goal of SNs is to collect the data from the area of interest and communicate it to the sink or base station (BS) for further processing via single or multi-hop transmission. Sometimes, the BS acts as a gateway to the Internet of Things (IoT), where the IoT can communicate the data to the Cloud using the Internet. The battery-equipped SNs consume more energy for heavy data transmission. Transmission of high-quality data in SN makes the battery-equipped micro-sensors consume much energy. Mobile Wireless Sensor Network (MWSN) represents a fast-evolving technology, and its use in many things is not limited. While fixed-infrastructure networks constrain sensor nodes to one specific location, MWSNs allow the partial nodes or all nodes to move wherever they want and communicate between themselves, making the whole system more flexible. Furthermore, MWSNs can be compared with respect to GPS, Bluetooth Low Energy (BLE), and existing wireless sensor networks in aspects of extended network lifespan, energy saving, multiband functionality, and high targeting. Nevertheless, pathfinding in MWSNs is very challenging since the sensor nodes are mobile, low-cost devices that are timeconstrained, allowing limited resources to be used. On the mobile network, this unique frequency scheme creates extra difficulty in routing. In most monitoring applications, only partial nodes need to be moved in the network. Such nodes are called mobile agent sink nodes or sensor nodes. In the present work, the movement of only a few nodes is considered in MWSN. Energy consumption and network connectivity are two major issues in MWSNs. Several studies have been conducted to develop and propose suitable solutions for these problems. Many researchers are working to develop the best solutions due to the severe problems with energy consumption and network connectivity in mobile wireless sensor networks. To investigate network connectivity, this study introduces a new efficient technique that considers parameters like network stability, detection area, low energy consumption, etc. This approach guarantees network connectivity, communication sustainability, and the highest level of energy consumption optimization. This research investigates network connectivity issue and proposes two routing algorithms, namely Self-Organizing Maps based-Optimized Link State Routing (SOM-OSLR) and Deep Reinforcement Learning based-Optimized Link State Routing (DRL-OLSR) for MWSNs. Both algorithms undertake the relationship between sensor node deployment, communication radius, and detection area and suggest a new way to maintain communication while optimizing energy usage. I have evaluated both algorithms through simulations by considering various performance metrics such as connection probability, end-to-end delay, overhead, network throughput, and energy consumption. The network is analyzed for proposed routing and aggregation methods to analyze the performance. The simulation analysis is discussed under three scenarios. The first scenario undertakes ’no optimization,’ the second considers SOM-OLSR, and the third undertakes DRL-OLSR. The simulation results indicate that the SOM-OLSR performs better compared to the case with ’no routing’ optimization. Comparing DRL-OLSR and SOM-OLSR indicates that the former outperforms the latter in terms of low latency and high network lifetime. Specifically, the DRL-OLSR achieves a 47% higher throughput and 67% lower energy consumption compared to the SOM-OLSR. In addition, when compared to the ’No optimization’ condition, the DRL-OLSR achieves a notable 69.7% higher throughput and almost 89% lower energy consumption. These findings highlight the effectiveness of the DRL-OLSR approach in optimizing network performance and energy efficiency in wireless sensor networks. Similarly, data aggregation consistently reduces energy consumption across all scenarios, with up to 50% lower as compared to without data aggregation.