摘要
针对目前动态建模和数据驱动的方法对中长期流量预测的研究存在计算功率消耗大和准确率不够高,不足以预测医疗物资运输中交通流量的不确定性和复杂性的问题,文中把病毒传播用图神经网络提出了一个以解决物资调度时间序列预测问题建模。以NGSIM数据集IDM参数和HighD数据集等为研究对象,基于GCN基于节点的图分类或预测的方法删除了自连接,引入了权重矩阵,同时将输出层替换为对每个节点的功能独立运行的前馈层。这样可以更好地解耦特征提取和根据提取的特征进行的最终预测。实验表明,优化的模型更有效捕获全面的时空相关性,将预测误差降低了30%以上达到最佳基准模型,提升了医疗物资调度效率。
Aiming at the problems that the research on medium and long-term flow prediction by dynamic modeling and data-driven method has high computational power consumption and low accuracy at present,which is not enough to predict the uncertainty and complexity of traffic flow in the transportation of medical supplies,a graph neural network model for virus transmission is proposed to solve the problem of material scheduling time series prediction.Taking the IDM parameters of NGSIM data set and HighD data set as the research objects,the GCN based graph classification or prediction method eliminates the self connection,introduces the weight matrix,and replaces the output layer with the feed-forward layer which operates independently for each node’s function.In this way,the feature extraction and the final prediction based on the extracted features can be better decoupled.The experiment shows that the optimized model can capture the overall spatial-temporal correlation more effectively,reduce the prediction error by more than 30%to achieve the optimal benchmark model,and improve the efficiency of medical material scheduling.
作者
黄承宁
李娟
陈嘉政
HUANG Cheng-ning;LI Juan;CHEN Jia-zheng(Nanjing Tech University Pujiang Institute,Nanjing 211222,China)
出处
《计算机技术与发展》
2021年第9期202-207,共6页
Computer Technology and Development
基金
国家自然科学基金(61702229)
江苏省高等学校自然科学研究项目(18KJD520001)
南京工业大学浦江学院人才培养工程计划项目(njpji2019-2-01)。
关键词
智慧医疗
图神经网络
智能调度
数据驱动
问题建模
intelligent medical
graph neural network
intelligent scheduling
data driven:problem modeling
