Abstract
<title>Abstract</title> <p>The Industrial Internet of Things (IIoT), leveraging edge computing's proximity to data sources, enables real-time monitoring and improves operational efficiency, and has therefore attracted growing attention. However, the complex environments of oil and gas field stations, characterized by massive metallic storage tanks, cause severe Line-of-Sight (LoS) signal blockages, undermining data transmission stability. Prior studies have devoted limited attention to the systemic impact of such absolute geometric obstacles. In this work, we study obstacle-aware edge server deployment and task scheduling. To address non-differentiable geometric constraints and dynamic task loads, we design a two-stage framework named DOA-EDS. In the first stage, an Obstacle-Aware Adaptive Large-Neighborhood Search identifies high-quality obstacle-free deployment coordinates through adaptive destroy-and-repair operations. In the second stage, a feasibility-masked Proximal Policy Optimization agent enables real-time task scheduling. Simulation-based evaluations show that DOA-EDS maintains 95.2% network throughput and reduces average service latency by 77.4% relative to DQN-Joint under the highest evaluated load.</p>