Research Article
Hyperparameter Analysis of Temporal Graph Convolutional Network Model Applied to Traffic Prediction
@INPROCEEDINGS{10.1007/978-3-030-72792-5_53, author={Jing Huang and Lei Chen and Yuan An and Kailiang Zhang and Ping Cui}, title={Hyperparameter Analysis of Temporal Graph Convolutional Network Model Applied to Traffic Prediction}, proceedings={Simulation Tools and Techniques. 12th EAI International Conference, SIMUtools 2020, Guiyang, China, August 28-29, 2020, Proceedings, Part I}, proceedings_a={SIMUTOOLS}, year={2021}, month={4}, keywords={GCN Machine learning Hyperparameters optimization Traffic prediction}, doi={10.1007/978-3-030-72792-5_53} }- Jing Huang
Lei Chen
Yuan An
Kailiang Zhang
Ping Cui
Year: 2021
Hyperparameter Analysis of Temporal Graph Convolutional Network Model Applied to Traffic Prediction
SIMUTOOLS
Springer
DOI: 10.1007/978-3-030-72792-5_53
Abstract
GCN based on time and space is an essential part of smart city construction because it can capture the spatiotemporal dynamics and effectively analyze the traffic data to get the best prediction results. In the specific operation of the model, the adjustment and optimal selection of super parameters can make the model provide the best results, thus saving time, cost and computing power. When it comes to the prediction scenarios with low computational power and urgent demand, the existing super parameter search methods and optimization models lack efficiency and accuracy. Therefore, this paper proposes a super parameter search and optimization method based on cross validation, which can efficiently and accurately optimize the parameters, and select the best parameters by using the similarity between the learning and training errors corresponding to each super parameter To improve the prediction ability of the model. Through the verification of the actual data set, the model runs well, and can provide the best prediction results for the traffic flow and other scenarios dominated by spatiotemporal state.
