Chapter eight - Grid-based origin-destination matrix prediction: a deep learning method with vector graph transformation similarity loss function

Abstract

Deep learning is a branch of machine learning that allows computational models to learn representations of data with multiple levels of abstraction. It is widely used in tansportation field. The study of Origin-Destination (OD) matrix prediction is an important part of the transportation field. Many researchers focus on building prediction models to improve accuracy. In this study, the vector graph transformation loss function is adopted into the OD matrix prediction model to measure the transformation cost from an OD matrix to another. For appropriately used the loss function, a convolutional neural network-based model is developed to approximate the vector graph transformation loss function. To find which deep learning model-based vector graph transformation loss function is better, 7 models are developed with the same structure, but different kernel size of the first 2D convolution layer. The result shows that the proposed loss function has better performance compared with traditional loss functions.