Driving Time Prediction at Freeway Interchanges Using Artificial Neural Network and Particle Swarm Optimization

Abstract

Regarding the role of time in human life, the driving time can be a tangible criterion to represent the quality of flow in every traffic facility such as freeway interchanges. There are a lot of appreciable attempts which focused on determining travel time in long-length paths in the networks. But, there are not enough considerable studies on predicting the driving time in short-length paths such as the parts of an interchange. Besides, when the study area has to be changed, or when the purpose is to design a nonexistent interchange, predicting the driving time in diverse parts of an interchange cannot be carried out by direct measurement, and when there is information shortage on some hardly achievable traffic-based characteristics of interchanges, predicting the driving time by common procedures will be so hard or even impossible. Using simulation software usually needs considerable time and budget, too. In this paper, focus was on studying the short-length fragments of freeway interchanges to predict driving time by investigating ten real-world interchanges and more than 13,600 different simulated parts of interchanges. An artificial neural network (ANN)-based model and a particle swarm optimization (PSO)-based model were developed, and therefore, predicting driving time was possible just based on basic simple traffic and geometrical properties of the interchange. The results of testing, validating, and statistical analysis depicted a proper and precise development of the models. Although the PSO-based models have higher RMSE values than ANN-based models, the models were reliable enough to use for predicting the driving time in four parts of interchanges.