الملخص الإنجليزي
Abstract:
Vehicular ad-hoc network (VANET) is considered a promising solution in intelligent transport system (ITS) and smart cities (SC) to enhance road safety, traffic efficiency, driver, and passenger's comfort. Numerous studies have been conducted in this field to provide a reliable and bandwidth-efficient broadcast protocols for IEEE 802.11p VANET. The main problem that will be addressed in this thesis is the deficiency of the existing VANET broadcasting protocols and channel access schemes that leads to broadcast storm, message collisions, and consequently, a waste of channel bandwidth and decrease packet delivery ratio (PDR). VANET utilizes a traditional broadcasting protocol that employs flooding techniques to disseminate received messages to vehicles within communication range. Moreover, VANET uses a carrier sense multiple access with collision avoidance (CSMA/CA) to access channel resources, utilizing a slotted binary exponential backoff (BEB) mechanism. It is based on a random backoff value called contention window (CW) to access the channel. The CW value is doubled after the unsuccessful transmission and rest after successful transmission. The proper setting of CW value has a significant impact on the efficiency of VANET network.
This thesis proposes a fuzzy logic inference to address the issues of broadcast storms and high message collisions. The protocol intelligently selects the next hop relay vehicle, considering factors such distance, mobility, and received signal strength indication (RSSI).
The thesis has also proposed new and improved artificial intelligence (AI) based broadcasting protocols and channel access schemes to improve the performance of the BEB algorithm and enhance the overall performance of VANETs. Three deep reinforcement learning (DRL) algorithms are proposed, Q-learning, deep Q-network (DQN) and actor- critic. Q-learning is a simple and effective algorithm that can be used to learn the optimal action-value function for a given state. DQN is a more complex algorithm that uses a deep neural network to approximate the action-value function, and Actor-critie is a hybrid algorithm that combines elements of both Q-learning and policy-gradient methods. By using DRL, an intelligent learner can dynamically and independently control the CW value to achieve its objectives. To apply RL and DRL algorithms in VANET environment, a hybrid framework called NS3-gym is utilized.
C
The proposed algorithms were implemented, tested, and evaluated in NS3-Gym; fuzzy logic has significant improvement by reducing rebroadcast packets compared to the conventional approach. The proposed QL and DRL algorithms learns various VANET environments and enhance their performance compared to conventional channel access protocol.
The acquired new insights on the network performance of the proposed algorithms can offer precise guidelines for efficient designs of practical, reliable, fair V2V communication systems for sparse and dense environments. These results can potentially have a significant impact on a wide range of related VANET applications.
