Abstract:
Connected Autonomous Vehicles (CAVs) have been considered as a promising technology
that provides various smart transportation applications such as autonomous driving,
traffic flow prediction, smart authentication systems, and congestion control. These
smart applications have computation-intensive tasks which require the processing of a
large amount of data generated from various sensors attached to vehicles. The vehicle
having less availability of computational resources (referred as a client vehicle) sends
computation requests to its nearby roadside unit (RSU) for processing. There are two
computing platforms in vehicular networks - i) Vehicular edge computing: RSU acts as
an edge server and computes the tasks locally, ii) Vehicular fog computing: RSU offloads
computation-intensive tasks to the fog server for further processing. In both computing
platforms, one of the most important problems is to reduce the energy consumption of
renewable energy-powered RSUs in computing or offloading computation-intensive tasks.
In vehicular fog computing, there are two kinds of fog servers - static and dynamic fog
servers. Static fog servers are placed at fixed locations inside their corresponding fog zones.
Dynamic fog servers include moving or parked vehicles having high processing capabilities.
RSUs can be in contact with several fog servers (static and dynamic) at a time and offload
computation-intensive tasks to any fog server for processing. Hence, fog servers may have
uneven computation load due to varying offloading rates across the road segments which,
in turn, increases the latency in the request fulfillment.
In this thesis, we propose an energy-efficient scheduling technique across RSUs (ee-IRSA),
a federated deep Q-learning-based offloading technique (FedDOVe), and a mobility-aware
caching technique (MobiCache) to minimize the energy consumption across RSUs and
achieve uniformity in the distribution of computation load across fog servers. The
simulation results show that our proposed techniques reduce the energy consumption of
RSUs, and improve the load balancing across fog servers as compared to existing techniques.
