CPU–GPU–DPU Heterogeneous Computation Offloading and Resource Allocation Scheme in UAV-Assisted Edge Computing Network

With the rapid development of artificial intelligence and big data applications, traditional CPU-centric computing architectures are increasingly inadequate in addressing the demands of emerging technologies such as large models and multimodal fusion. The new generation of data-centric heterogeneous computing architectures, which coordinate the operations of DPU, CPU and GPU, has emerged as a crucial technology for overcoming performance bottlenecks associated with the exponential increase in data traffic. In this paper, we first propose a Four-Stage Heterogeneous Computing (FSHC) model to effectively describe the heterogeneous computing process of CPU-GPU-DPU collaboration. Considering multiple UAV-enabled aerial computing (MUEAC), heterogeneous computing resources, and cloud-edge-end collaboration, the joint optimization problem is formulated to minimize energy consumption and delay of the system. To tackle this challenging problem, we propose a two-layer optimization algorithm combining Self-Organizing Map and chaotic local PID-based search algorithm (ToSMPSA), which utilizes Self-Organizing Map (SOM) in the upper layer to solve terminal area division and UAV deployment planning, and employs chaotic local PID-based search algorithm (CLPSA) in the lower layer to solve task offloading decisions and heterogeneous computing resource allocation strategies. Simulation results demonstrate that the proposed method significantly outperforms benchmark schemes in reducing system overhead.