Intelligent Resource Allocation and Partial Computation Offloading in Mobile Edge Computing Using Duelling Multi-Branch Duelling Deep Q Networ

Abstract

The rapid expansion of marine activities has significantly increased the demand for
computation-intensive applications in the Maritime Internet of Things (M-IoT), creating
challenges in delivering reliable and efficient computing services. Computation offloading has
emerged as an effective approach for resource-constrained IoT devices by transferring intensive
tasks to edge servers, thereby reducing latency and energy consumption. However, conventional
Deep Reinforcement Learning (DRL) techniques often struggle with high-dimensional state and
action spaces, leading to suboptimal offloading decisions. To address these limitations, this paper
proposes a Multi-Branch Duelling Deep Q Network (MBDDQN) for efficient computation
offloading in dynamic marine environments. The proposed architecture decomposes complex
decision-making into multiple branches, reducing action space complexity and enhancing
scalability. Additionally, a Long Short-Term Memory (LSTM) network is integrated to capture
temporal dependencies and improve long-term reward estimation. An adaptive resource
allocation mechanism further balances communication and computation costs, minimizing delay
and energy consumption. Extensive simulation results demonstrate that the proposed MBDDQN
outperforms conventional methods such as DQN, Double DQN, and Dueling DQN. The
achieved F1-score of 98.83% further confirms the model’s robustness, ensuring a strong balance
between precision and recall for reliable and efficient computation offloading in M-IoT
environments.