P.W. Deng, H.J. Dai, Z. Lou, C.Y. Dong

Pages: 177-190

Abstract
To improve the low traffic flow efficiency on highways, this study focuses on digital infrastructure construction, specifically on real-time traffic flow prediction. A multi-level framework for digital traffic infrastructure is designed and implemented. Various sensors are utilized for data acquisition. The framework involves preprocessing to eliminate redundant information, and data management through a database system. Prediction analysis is conducted using a combination of wavelet neural networks and a Shuffled Frog Leaping algorithm. The Shuffled Frog Leaping short-term traffic flow prediction algorithm and radial basis function neural networks are employed for feature extraction and model optimization. In the experiment, the traffic flow prediction model developed in this study was compared with various advanced models for performance evaluation. The average absolute error ranged from 1.6235 to 2.7482, with a mean of approximately 1.9793. The average absolute error increased by 28.53%, the average absolute percentage error decreased by about 45.8%, and the root mean square error improved by nearly 26.9%. In 15 experiments, the predictive performance showed significant enhancements in error control, fitting degree, stability, and accuracy, demonstrating the superiority and practicality of the traffic flow prediction.
Keywords: highway; digital construction; traffic flow; smart transportation; Shuffled Frog Leaping