Research on Edge Computing Deep Neural Network Task Unloading Based on Resource Collaboration Framework and Multi Strategy Optimization

In the era of the Internet of Things, the exponential growth of IoT devices and explosive expansion of data have driven deep neural networks (DNNs) to become the core of intelligent data processing.However, their high computational complexity conflicts with the performance and energy consumption limitations of terminal devices.Existing schemes have limitations such as high hardware cost, lightweight model accuracy loss, high cloud computing latency, and lack of single path collaboration in edge computing.This research proposes the edge computing resource collaboration framework CoDNN, which optimizes throughput by executing DNN tasks through multiple computing resource pipelines; Design a particle swarm optimization based partition offloading algorithm DBPSO for single task delay optimization, dynamically allocating resources to meet deadline constraints; Aiming at the scenario of multi terminal sharing heterogeneous resources, a multi task energy consumption optimization strategy and the Levy flight particle swarm algorithm DBLPSO-MD are proposed to synchronously achieve deadline constraints and minimize system energy consumption.Research has found that the DBPSO algorithm has better response time than existing strategies, and DBLPSO-MD has significant advantages in achieving deadlines and energy consumption when resources or tasks increase.Through resource collaboration framework and multi strategy optimization, this research has promoted the efficient execution and energy efficiency improvement of DNN tasks in edge computing scenarios, and provided an effective solution for intelligent applications of the Internet of Things.