基于GPU的一类地理多智能体系统并行仿真研究

GPU-Accelerated Simulation for Class of Multi-Agent Based Models

针对当前地理多智能体建模存在着计算成本高、配置复杂、运算加速性能不高的问题,以杜能模型为例,提出基于GPU并行技术的一类地理多智能体仿真与优化方法。通过构建空间索引网格的方法,动态维持智能体与空间索引网格的关联关系,提高地理多智能体系统的仿真运行效率。研究结果表明:采用GPU并行技术,能够使多智能体系统的运行性能得到明显提升,对开展大规模数据下的空间系统多智能体仿真建模具有重要意义。

A parallel agent-based model of Von Thünen Model was proposed driven by graphics processing units(GPUs). The Von Thünen Model often involved the simulation of large numbers of geographically located individual decision-makers and a massive number of individual-level interactions. This simulation required substantial computational power. GPU-enabled computing resources provided a massively parallel processing platform based on a fine-grained shared memory paradigm. This massively parallel processing platform held considerable promise for meeting the computing requirement of agent-based models of spatial problems. A dynamic relationship table rebuilding method was proposed to enable the use of GPUs for parallel agent-based modeling of the spatial Von Thünen Model. The key algorithm played an important role in best exploiting high-performance resources in GPUs for large-scale spatial simulation. Experiments conducted to examine computing performance show that GPUs provide a computationally efficient alternative to traditional parallel computing architectures and substantially accelerate agent-based models in large-scale spatial space.