Konu "Kent ulaşımı -- Türkiye -- Eskişehir" ile LEE- Deniz Ulaştırma Mühendisliği-Doktora'a göz atma
Sayfa başına sonuç
ÖgeModelling departure time, destination and travel mode choices by using the generalized nested logit model: an example for discretionary trips( 2020) Elmorssy, Mahmoud Morssy Mohamed ; Tezcan, Hüseyin Onur ; 635767 ; İnşaat Mühendisliği Ana Bilim DalıNowadays, understanding the influences of different temporal and spatial factors on individuals' travel choices becomes essential especially after the pandemic of COVID-19 that invaded the world in 2020. Such an outbreak had its own influences on the future transportation planning studies. By words, policy makers have directed their interests toward newly emergency transportation policies that aim to distribute travels over wider time and space spans in accordance with precautionary and preventive measures to counteract Corona virus or any other similar future virus attacks. However, transportation planning studies still rely on traditional demand modelling approaches such as the four-step model. The four-step model is still exposed to considerable criticism for its shortages in representing the potential correlations between temporal, spatial factors and different travel dimensions which leads to inaccurate representations of individuals' actual travel behaviour. In order to overcome that, some researches have directed their interests toward using choice modelling approach as an alternative to some stages in four-step model. Even though these approaches show better performance in terms of goodness of fit and predictability power, most of them have represented travel dimensions individually rather than jointly. As there is a gap in literature about representing a unified choice model that connect different travel demand dimensions and consider various potential inter-correlation among them, this dissertation contributes filling this gap through introducing three research papers that employ various types of discrete choice models for jointly representing three major travel dimensions; destination, departure time and travel mode. Such models contribute more to mathematical modelling literature of transportation demand models that provide more detailed and specific micro-policy analyses where traditional four-step model cannot. The presented papers introduce three discrete choice models that differ in the level of accounting for correlation of error terms within elementary alternatives and therefore differ in cross-elasticity pattern while offering computational simplicity. In the first paper, limited number of correlation patterns is introduced by adopting the three-level Nested Logit (NL) models. In the second paper, opposite to traditional NL models that was introduced in previous paper, this paper assesses the effect of considering spatial correlation of adjacent discretionary destinations on the choice of the two other travel dimensions by using the Ordered Generalized Extreme Value (OGEV) approach. The third paper, introduces a novel modelling methodology for using the Generalized Nested Logit (GNL) model to represent multi-dimensional potential correlations; between different travel dimensions (inter-correlation), inside the same travel dimension (inner-correlation) and correlation due to ordered nature travel dimensions (e.g. spatial correlation among destinations and temporal correlation between departure times). Overall, in the published papers, different levels of correlation between departure time, destination and travel mode choices and within each travel dimension are represented through different assumed correlation structures according to the nesting structure limitations provided by each model. Moreover, the associated formulas for each proposed model that reflect different patterns of correlation (cross-elasticity) are explicitly introduced. From a policy implications standpoint, a calibrated version of departure time, destination and travel mode model will provide policy makers very detailed analyses about the inter-relationships associated with the three travel dimensions (while traditional four-step model cannot provide at micro-level). That leads to more certain, specific, efficient and precise policy decisions. Thus, developing these models can be considered as a significant milestone toward obtaining a consistent, efficient and integrated full-scale model that can lie in all travel demand dimensions (e.g. number and duration of activities for activity and tour-based models). The developed models have been estimated and calibrated by using shopping and entertainment trips data of Eskisehir city, Turkey. The data have been collected through a household survey that was conducted in 2015 in the context of Eskisehir strategic master plan project which was operated by Eskisehir Metropolitan Municipality. Eskisehir is a city in north-western Turkey. It is considered as a medium sized city with a population of 799724 (2013 census) distributed over about 2678 km2 area. The collected data include variables that represent attributes of alternatives and individuals' characteristics to be used in models' utility functions. The first group of alternatives' attributes is travel time related attributes where, in vehicle time and out of vehicle time (egress time, at stop waiting time and access time) for each individual trip have been obtained. Moreover, related to travel cost, the fare of public transportation modes (for public transportation users), trip cost for private cars as well as parking fees (for private car users) have been observed for each individual trip. Within the collected revealed data, a good portion of socio-economic individual characteristics related observations are presented. These data include car ownership, individual's age, monthly income and student status (if respondent is a student or not). The total number of observations related to the determined alternatives has been found to be 529. The estimation results of each model have been explicitly interpreted in each paper and logical as well as statistical comparisons between pairs of models have been conducted in order to ensure the superiority of more advanced approaches (OGEV and GNL) over the lesser ones (NL). In the light of the estimation results, generally, individuals have been found to jointly decide on "at which departure time", "to which destination" and "by which mode" rather than doing this separately as assumed by traditional four-step model. Neglecting the potential correlation among alternatives of the three travel dimensions has led to inaccurate estimates of measurements' indicators such as Value of Time (VOT) which results finally in incorrect and improper policy decisions. From another hand gradual improvements in predictability have been observed as the level of the represented correlation increases. That is, three-level NL model was found to offer improvements over Multinomial Logit (MNL) model, OGEV model is prominent over NL model and GNL is superior over all models. It is possible to argue that the proposed GNL approach has distinct improvements over all other proposed approaches. Its simplicity along with the incomparable flexibility in representing a lot of correlation patterns within and among three vital travel dimensions all of that under a unified modest model qualify it to be prominent. The proposed GNL model has provided very detailed analyses about the inter-dependencies associated with various departure times, travel modes and discretionary destinations where other models cannot. The estimation results have expressed the powerful analytical ability of the proposed GNL approach where it has the power of capturing unusual correlation patterns. These patterns are thoroughly specific, unexpected, and very difficult to be observed in the market. By words, the dissertation argues that there is no other approach as simple as the proposed GNL and leads to such temporal and spatial specific analyses. The advantages associated with the proposed GNL approach qualify it to be a strong peer to the traditional four-step model in micro-disaggregate modelling scopes if applied for medium and small-scale planning studies that involve limited number of alternatives in each travel dimension. It may be used with a large number of alternatives in each travel dimensions as well, however, through stratifying the whole population to small segments based on one or more travel dimensions to produce small segments suitable for readily estimation process. Finally, the proposed GNL methodology represents a time of day-based trip-end distribution model that can reproduce a considerably more accurate transportation mode based origin-destination matrix dependent on time of day. Moreover, unlike traditional four-step models, parameter estimates produced from the GNL model can provide significant indications which precisely reflect the individuals' actual behaviour. Obviously, that can enormously help policy makers to reach a solid perception about the effects of applying various strategies to manage demands through different times of day and towards different destinations.