Deep reinforcement learning based offloading decision algorithm for vehicular edge computing.

Area of Science:

• Vehicular Edge Computing (VEC)
• Artificial Intelligence
• Distributed Systems

Background:

• Task offloading is crucial for VEC, balancing time, energy, and performance.
• Traditional methods lack optimal Mobile Edge Computing (MEC) server utilization and ignore vehicle mobility.
• Signal blocking is a common challenge in VEC environments.

Purpose of the Study:

• To propose a novel algorithm for efficient task offloading decisions in VEC.
• To address the limitations of traditional distributed task offloading methods.
• To minimize system overhead by considering vehicle mobility and signal blocking.

Main Methods:

• Developed Deep Reinforcement learning based offloading decision (DROD) algorithm for VEC.
• Modeled vehicle-MEC interactions using a Markov decision process.
• Employed an improved deep deterministic policy gradient (NLDDPG) with LSTM for training.

Main Results:

• DROD effectively considers vehicle mobility and signal blocking in its decision-making.
• The NLDDPG algorithm, with normalized state space and LSTM, enhances learning efficiency.
• Experimental results show DROD outperforms DQN (25%), NLDQN (10%), and DDDPG (130%).

Conclusions:

• DROD offers a superior approach to task offloading in VEC compared to existing algorithms.
• The proposed method enhances resource utilization and system performance in dynamic VEC environments.
• Considering mobility and signal blocking is vital for optimizing VEC task offloading.