Edge Computing Optimization Framework for Real-Time IoT Data Processing: A Reinforcement Learning-Based Approach

The rapid rise in IoT real-time data processing demands challenges traditional cloud computing, limited by high latency and bandwidth bottlenecks in applications like intelligent transportation. This paper introduces an edge computing optimization framework to improve real-time performance, resource efficiency, and adaptability. Targeting intelligent transportation, the framework employs a multi-layer architecture (sensor, edge, cloud) with three key components: a data preprocessing module reducing transmission overhead via filtering and compression, a reinforcement learning-based task scheduling algorithm (RL-TS) dynamically optimizing resources using state space (node load, task demands), action space (task allocation), and a reward function balancing latency, throughput, and energy, and a communication optimization module minimizing redundancy through deduplication and incremental updates. Evaluated with Caltrans PeMS and synthetic datasets, RL-TS outperforms baselines (Cloud-Only, Greedy-Edge, LB-Edge) across edge node counts (5-20) and loads (5-20 tasks/s). It achieves an average latency of 152 ms (68.7% below Cloud-Only’s 485 ms), throughput of 16.5 tasks/s (15.4%-101% higher), and energy use of 450 J (42.3% less than Cloud-Only’s 780 J). RL-TS demonstrates superior robustness and scalability, providing an efficient solution for IoT edge computing, with future potential in algorithm and cross-scenario enhancements.