AGIR - A decision support tool for Geneva ring-road traffic management during important events

Geneva is a city used to organise international events like for example the International Motor Show which takes place every year. Most of these events are hold on the Palexpo area, a group of buildings situated near the Geneva airport. Both places are closely connected to the city ring highway. Traffic generated by this type of events obviously implies a totally different demand distribution shape compared to usual days. In addition, traffic volumes recorded during the event peak hours, (usually observed around the opening and closing hours), can reach up to the double of the ones ordinarily experienced. Given that the road network has not been design to handle this type of demand, important congestions are consequently observed during event periods implying important accidents risk, parking lot saturation and an overuse of the local network provoking annoyances for residents. In order to minimize these problems, the Geneva Transport Office asked Arx iT and CITEC (both local consultancy companies) to develop a method for supporting traffic management on the ring road. The basic concept is to implement in the Geneva highway traffic centre an on- line traffic data analysis tool which is able to, depending on the traffic situation (data coming from loop detectors and cameras) and parking lots occupancy, suggest a set of recommendations to be provided to road users through variable message signs. The whole process mainly relies on comparisons between the current network status and historical situations already observed during past events. For each of these situations correspond a traveller information scenario which has been designed based on empirical experiences. Although this tool has been offering reasonably good results for three years, situations without clear correspondence with past conditions are still frequently observed and the lack of experience on this particular cases lead sometimes to erroneous information scenario selection. Of course, facing these new situations allows traffic operators to increase their human knowledge, on which mainly relies the current AGIR process. Nevertheless, need for more systematic analysis process has been clearly identified. Reproducing various traffic situations without having to wait them to occur in reality is one of the major purposes of using traffic simulation tools in transportation studies. Thus, EPFL (Swiss federal Institute of Technology) has been in charge to assess the current AGIR system performances and to improve the knowledge data base through artificial experiments using a microsimulation model of the Geneva ring highway and part of the local network. For this task, EPFL has been helped by TSS, the company developing AIMSUN NG, the traffic simulator used for this analysis. The use of microsimulation offered the possibility of dynamically study the network’s traffic reaction to the information provided to simulated drivers by the AGIR system. This reaction implying new traffic data outputs, which are used as input for the AGIR process, the time dependent interaction between traffic situation and information system is fully captured. As simulation models offer statistical traffic outputs for each network element, global performance indexes could be computed offering a more realistic assessment of situation that only relying on traffic detection data. This complementary approach has been shown to be particularly efficient to improve current process and offer better answers to non frequent situations. Currently done off- line, possible integration of the simulation tool for on- line decision process is discussed for the future.

Methodology for traffic risks identification

Edward Chung, André-Gilles Dumont, Minh Hai Pham

Of all modes of transport, transport by road is the most dangerous and the most costly in terms of human lives, according to Road Safety Action Program (2003-2010). In Europe, the 2010 road safety target of halving the number of road fatalities by 2010 seems difficult to be obtained although road safety has been improved recently. Focusing on crash prevention on motorways, this paper proposes a framework for identifying risky traffic conditions using logistic regression models for specific traffic conditions. The innovative feature of this framework is that we evaluate the traffic risk under traffic regimes obtained by Self-Organizing Maps (SOM). We aggregate traffic data from double loop detectors and weather data from weather stations to produce traffic status for 5-minute intervals that we call traffic situations. Using Principal Component Analysis (PCA), we transform the traffic situations into new data space ready for the clustering process by SOM. The PCA transformation allows reducing the number of dimensions and removing random noise in the data. The clustering process by SOM results in groups of similar traffic situations that we call traffic regimes. From crash database, we determine pre-crash traffic situations, which precede crash occurrences. With the obtained clustering results, we classify pre-crash situations into traffic regimes. Under each traffic regime, we develop a logistic regression model for identifying risky traffic situations, which is specific for that traffic regime. The promising result shows that the models can identify correctly high percentage of risky situations while the rate of false detection is low.

2009

Multivariate time series forecasting for freeway networks

Semin Kwak

This dissertation introduces traffic forecasting methods for different network configurations and data availability.Chapter 2 focuses on single freeway cases.Although its topology is simple, the non-linearity of traffic features makes this prediction still a challenging task.We propose the dynamic linear model (DLM) to approximate the non-linear traffic features. Unlike a static linear regression model, the DLM assumes that its parameters change over time.We design the DLM with time-dependent model parameters to describe the spatiotemporal characteristics of time-series traffic data. Based on our DLM and its model parameters analytically trained using historical data, we suggest the optimal linear predictor in the minimum mean square error (MMSE) sense.We compare our prediction accuracy by estimating expected travel time based on the traffic prediction for freeways in California (I210-E and I5-S) under highly congested traffic conditions with other baselines. We show significant improvements in accuracy, especially for short-term prediction.Chapter 3 aims to generalize the DLM to extensive freeway networks with more complex topologies.Most resources would be consumed to estimate unnecessary spatiotemporal correlations if the DLM was directly used for a large-scale network.Defining features on graphs relaxes such issues by cutting unnecessary connections in advance based on predefined topology information.Exploiting the graph signal processing, we represent traffic dynamics over freeway networks using multiple graph heat diffusion kernels and integrate the kernels into DLM with Bayes' rule. We optimize the model parameters using Bayesian inference to minimize the prediction errors.The proposed model demonstrates prediction accuracy comparable to state-of-the-art deep neural networks with lower computational effort. It notably achieves excellent performance for long-term prediction through the inheritance of periodicity modeling in DLM.Chapter 4 proposes a deep neural network model to predict traffic features on large-scale freeway networks.These days, deep learning methods have heavily tackled traffic forecasting problems of freeway networks because they are outstanding at learning highly complex correlations between variables both in time and space, which the linear models might be limited to.Adopting a graph convolutional network (GCN) becomes a standard to extract spatial correlations; therefore, most works have achieved great prediction accuracy by implanting it into their architecture.However, the conventional GCN has the drawback that receptive field size should be small, i.e., barely refers to traffic features of remote sensors, resulting in inaccurate long-term prediction.We suggest a forecasting model called two-level resolution deep neural network (TwoResNet) that overcomes the limitation.It consists of two resolution blocks:The low-resolution block predicts traffic on a macroscopic scale, such as regional traffic changes.On the other hand, the high-resolution block predicts traffic on a microscopic scale by using GCN to extract spatial correlations, referring to the regional changes produced by the low-resolution block.This process allows the GCN to refer to the traffic features from remote sensors.As a result, TwoResNet achieves competitive prediction accuracy compared to state-of-the-art methods, especially showing excellent performance for long-term predictions.

EPFL2022

Multivariate time series forecasting for freeway networks

Semin Kwak

EPFL2022