基于生成对抗网络的时空交通数据预测方法研究  

Research on Spatio-temporal Traffic Data Prediction Method Based on Generative Adversarial Network

作　　者：郑春晖 ZHENG Chunhui(Department of Information Engineering,Yantai Gold College,Yantai 265401,China)

机构地区：[1]烟台黄金职业学院,信息工程系,山东烟台265401

出　　处：《微型电脑应用》2024年第9期54-55,64,共3页Microcomputer Applications

基　　金：山东省社会科学规划研究项目(20CPYJ03)。

摘　　要：交通流预测是时空数据挖掘领域的一个核心研究方向,其挑战在于交通流的高度非线性和复杂模式使得预测结果难以满足实际需求。现有方法往往忽略了时空动态性建模以及整合外部环境因素的重要性,因此,提出一种基于生成对抗网络(GAN)的时空图交通流预测方法,通过无监督学习时空图特征表示,实现更精确的时空序列图预测。所提出的方法采用GAN结构,通过博弈学习数据的分布并进行变分推理。实验结果表明,基于GAN的时空图交通预测方法在中国台湾高速交通数据集上的表现优于基线方法,能够更有效地解决预测结果平滑性的问题,获得更高效和准确的预测结果。Traffic flow prediction represents a crucial research avenue within the domain of spatio-temporal data mining.However,prevailing predictions fall short of expectations due to the highly nonlinear and intricate nature of traffic patterns.Existing methodologies often overlook the significance of capturing spatial-temporal dynamics and integrating external environmental variables.A novel approach for traffic flow prediction termed as the spatial-temporal graph traffic flow prediction method,leveraging generative adversarial network(GAN).This approach entails unsupervised learning of spatial-temporal graph feature representations to facilitate more precise spatial-temporal sequence graph prediction.Utilizing the GAN framework,the method engages in game-based learning to model data distributions and employs variational inference.Extensive experimentation on Chinese Taiwan’s high-speed traffic dataset validates the efficacy of the proposed method,demonstrating its superior capability in addressing the issue of prediction result smoothing compared to baseline approaches,obtaining more efficient and accurate forecasting results.

关 键 词：生成对抗网络 交通流量 时空图 图结构生成 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]