10:00 AM - 10:20 AM
[121] Subrogate Modeling for Network Design and Manifold Learning for Route Choice Model and OD Matrix with Origin and Destination Data as Latent Variables
Keywords:Subrogate Model, Manifold, Feedforward Neural Network
This paper focuses on pedestrian behavior and attempts to construct a method to simultaneously estimate the Origin-Destination traffic distribution and the route choice models. For the objective function, we propose an estimation algorithm that guarantees the uniqueness of the convergent solution from the viewpoint of information geometry. Based on the proposed method, the estimation is carried out using measured data.
Using a proxy model, we also attempt to accelerate the network design problem of urban functions from a quantitative point of view, such as tourists' expected utility. However, since the solution set of the network design problem is discrete, and a recursive logit model computes the assignment, the computational cost is enormous. Therefore, we propose a framework for speeding up the problem by training a feed-forward neural network using supervisory data of network and objective function values.
Using a proxy model, we also attempt to accelerate the network design problem of urban functions from a quantitative point of view, such as tourists' expected utility. However, since the solution set of the network design problem is discrete, and a recursive logit model computes the assignment, the computational cost is enormous. Therefore, we propose a framework for speeding up the problem by training a feed-forward neural network using supervisory data of network and objective function values.