To have a comprehensive overview of the current challenges in the transportation sector, it is essential to analyze traffic behavior and, particularly, to estimate the demand realistically. Therefore, traffic demand, represented by origin-destination (OD) matrices, is a vital key input for many traffic-related applications in traffic planning and management domains. Many studies have developed models to estimate single-modal traffic demand matrices. The conventional models use section traffic counts and traffic surveys as inputs. Unfortunately, it is highly expensive and time-consuming to carry out traffic survey campaigns as the process is not fully automated. A few studies have also developed models to estimate the traffic demand of multimodal shared systems, for example, public or freight transportation systems. However, these models depend on rich data provided by particular data sources, such as user smart cards of public transport systems. To the best of our knowledge, there are no models for estimating multimodal traffic demand of private transportation modes, such as driving, cycling, and walking. We argue that the significant hurdle to developing such models is the lack of reliable data. GPS modules allow the automatic collection of floating data (FD). FD are extensive trip records that provide location coordinates, timestamps, and speed values of devices equipped with active GPS modules. This kind of data can provide the required rich information and significantly reduce the disadvantages of traffic survey campaigns. This work aims to estimate multimodal traffic demand matrices of private transportation modes by fusing different data sources. Specifically, it develops a model to estimate OD matrices of driving, cycling, and walking using traffic counts and FD. To achieve this, the work is divided into three main parts. In the first part, we exploited the potential of floating car data (FCD) to enhance the quality and performance of the demand estimation process for vehicles. This was done by improving all input data of the information minimization (IM) model using FCD. The output of the proposed model was compared to the conventional bi-level demand estimation model. The results confirm that FCD improves the process efficiency and estimation quality. To estimate multimodal OD matrices, floating smartphone data (FSD) should replace FCD in the proposed model. The major disadvantage of automatically collected FSD is the missing information about the used transportation modes for conducting the trips. Many methods in the litera-ture rely on supervised machine learning (ML) algorithms to tackle this problem. However, such algorithms require labeled data, which are not always available. This research part aimed to search for a reliable method to infer transportation modes from unlabeled data. It did this by benchmarking different unsupervised and supervised ML algorithms with various input attributes. Two unsupervised algorithms proved accurate enough when using a reliable data attribute as a feature. The objective of the last research part was to investigate the ability of the developed model in the first part to estimate multimodal OD matrices using the processed FSD in the second part. The main challenge was adjusting the model to consider different transportation modes and obtaining input and validation data for all three modes. Therefore, we launched a traffic surveillance campaign alongside the simulation study to collect real counting and ground truth data using video cameras. These data were used for the field study. Furthermore, we conducted a sensitivity analysis using synthetic data to reflect the development of the estimation accuracy by increasing the penetration rate of FSD in a simulation environment. This part found the proposed model estimated the demand of the car and bicycle modes determined in the field study with an average correlation coefficient of 92%. This result corresponds to the simulation study's sensitivity analysis, which indicates that the model should achieve a demand estimation with an average correlation coefficient of 93% to 96%. For future work, we recommend further developing the model to rely only on FSD as one data source so the need for section traffic counts is no longer a prerequisite.


    Access

    Download

    Check availability in my library


    Export, share and cite



    Title :

    A Data Driven Approach for Estimating Traffic Demand of Different Transportation Modes


    Contributors:

    Publication date :

    2023


    Remarks:

    Schriftenreihe des Instituts für Verkehr und Stadtbauwesen, vol. 66



    Type of media :

    Miscellaneous


    Type of material :

    Electronic Resource


    Language :

    English



    Classification :

    DDC:    388



    A data driven approach for estimating traffic demand of different transportation modes

    Dabbas, Hekmat / Technische Universität Braunschweig | TIBKAT | 2023


    A data driven approach for estimating traffic demand of different transportation modes

    Dabbas, Hekmat / Technische Universität Braunschweig | TIBKAT | 2023

    Free access


    Estimating Transportation Demand

    Sinha, Kumares C. / Labi, Samuel | Wiley | 2007


    Estimating Service-Related Traffic Demand from Trip Chain Data

    Schneider, Sebastian | Online Contents | 2013