京都大学 / 0048 / 新制・課程博士 / 博士(工学) / 甲第22653号 / 工博第4737号 / 新制||工||1740(附属図書館) / 京都大学大学院工学研究科都市社会工学専攻 / (主査)教授 山田 忠史, 教授 藤井 聡, 准教授 SCHMOECKER Jan-Dirk / 学位規則第4条第1項該当 / Doctor of Philosophy (Engineering) / Kyoto University / DFAM

The conventional methods of determining the Origin-Destination (O-D) trip tables involve elaborate surveys, such as home interviews, requiring considerable time, manpower and funds. To overcome this drawback, a number of theoretical models that synthesize O-D trip tables from link volume data have been developed in the past. The focus of this research was on two of these models, namely, The Highway Emulator (THE) and the Linear Programming (LP) models. These models use target/seed tables for guiding the modeled trip tables. In an earlier research effort conducted by Virginia Tech Center for Transportation Research, potential was noted for enhancing these models' performances by using a superior target/seed table. This research study exploits the readily available socio-economic/census data and link volume information and proposes a methodology for obtaining improved target/seed tables, by performing the trip generation and trip distribution steps. This table was provided as target to THE and LP models, and their performances evaluated using Pulaski town as case study. In addition to measuring the closeness of the output tables to surveyed tables and their capability to replicate observed volumes, their improvements over the case when a structural table is used as target was also studied. Tests showed that the use of the superior target/seed table significantly improved the performance of the LP model. However, for THE, mixed trends are seen in terms of different measures of closeness. The sensitivity of the user parameter to place certain degree of belief on the target/seed table for LP model was also analyzed. / Master of Science

Courier companies receive delivery orders at different times in advance. Some orders are known long beforehand, some arise with a very short notice. Currently the order delegation, deciding which car is going to drive which order, is performed completely manually by a (TL) where the TL use their experience to guess upcoming orders. If delivery orders could be predicted beforehand, algorithms could create suggestions for vehicle routing and vehicle selection. This thesis used the data set from a Stockholm based courier company. The Stockholm area was divided into zones using agglomerative clustering and K-Means, where the zones were used to group deliveries into time-sliced Origin Destination (OD) matrices. One cell in one OD-matrix contained the number of deliveries from one zone to another during one hour. Long-Short Term Memory (LSTM) Recurrent Neural Networks were used for the prediction. The training features consisted of prior OD-matrices, week day, hour of day, month, precipitation, and the air temperature. The LSTM based approach performed better than the baseline, the Mean Squared Error was reduced from 1.1092 to 0.07705 and the F1 score increased from 41% to 52%. All features except for the precipitation and air temperature contributed noticeably to the prediction power. The result indicates that it is possible to predict some future delivery orders, but that many are random and are independent from prior deliveries. Letting the model train on data as it is observed would likely boost the predictive power. / Budföretag får in leveransorder olika tid i förväg. Vissa order är kända lång tid i förväg, medan andra uppkommer med kort varsel. I dagsläget genomförs orderdelegationen, delegering av vilken bil som kör vilken order, manuellt av en transportledare (TL) där TL använder sin erfarenhet för att gissa framtida order. Om leveransorder skulle kunna förutsägas i förväg kan fordonsrutter och fordonsval föreslås av algoritmer. Denna uppsats använde sig utav ett dataset från ett Stockholmsbaserat budföretag. Stockholmsområdet delades in i zoner med agglomerativ klustring och K-Means, där zoner användes för att gruppera leveranser in i tidsdelade Ursprungsdestinationsmatriser (OD-matriser).  En cell i en OD-matris innehåller antalet leveranser från en zon till en annan under en timme. Neurala nätverk med lång-kortsiktsminne (LSTM) användes för förutsägelsen. Modellen tränades på tidigare OD-matriser, veckodag, timme, månad, nederbörd, och lufttemperatur. Det LSTM-baserade tillvägagångssättet presterade bättre än baslinjen, det genomsnittliga kvadratfelet minskade från 1,1092 till 0,07705 och F1-poängen ökade från 41% till 52%. Nederbörd och lufttemperatur bidrog inte märkbart till förutsägelsens prestation. Resultatet indikerar att det är möjligt att förutse vissa leveransorder, men att en stor andel är slumpmässiga och oberoende från tidigare leveranser. Att låta modellen tränas med nya data när den observeras skulle troligtvis öka prognosförmågan.

Origin-Destination (O-D) trip tables represent the demand-supply information of each directed zonal-pair in a given region during a given period of time. The effort of this research is to develop a linear programming methodology for estimating O-D trip tables based on observed link volumes. In order to emphasize the nature of uncertainty in the data and in the problem, the developed model permits the user's knowledge of path travel time to vary within a band-width of values, and accordingly modifies the user-optimality principle. The data on the observed flows might also not be complete and need not be perfectly matched. In addition, a prior trip table could also be specified in order to guide the updating process via the model solution. To avoid excessive computational demands required by a total numeration of all possible paths between each O-D pair, a Column Generation Algorithm (CGA) is adopted to exploit the special structures of the model. Based on the known capacity of each link, a simple formula is suggested to calculate the cost for the links having unknown volumes. An indexed cost is introduced to avoid the consideration of unnecessary passing-through-zone paths, and an algorithm for solving the corresponding minimum-cost-path problem is developed. General principles on the design of an object-oriented code are presented, and some useful programming techniques are suggested for this special problem. Some test results on the related models are presented and compared, and different sensitivity analyses are performed based on different scenarios. Finally, several research topics are recommended for future research. / Master of Science

This research contributes to a very active body of literature on the application of Automated Data Collection Systems (ADCS) and openly shared data to public transportation planning. It also addresses the interaction between transit demand and land use patterns, a key component of generating time-varying origin-destination (O-D) matrices at a route level. An origin-destination (O-D) matrix describes the travel demand between two different locations and is indispensable information for most transportation applications, from strategic planning to traffic control and management. A transit passenger's O-D pair at the route level simply indicates the origin and destination stop along the considered route. Observing existing land use types (e.g., residential, commercial, institutional) within the catchment area of each stop can help in identifying existing transit demand at any given time or over time. The proposed research addresses incorporation of an alighting probability matrix (APM) - tabulating the probabilities that a passenger alights at stops downstream of the boarding at a specified stop - into a time-varying O-D estimation process, based on the passenger's trip purpose or activity locations represented by the interactions between transit demand and land use patterns. In order to examine these interactions, this research also uses a much larger dataset that has been automatically collected from various electronic technologies: Automated Fare Collection (AFC) systems and Automated Passenger Counter (APC) systems, in conjunction with other readily available data such as Google's General Transit Feed Specification (GTFS) and parcel-level land use data. The large and highly detailed datasets have the capability of rectifying limitations of manual data collection (e.g., on-board survey) as well as enhancing any existing decision-making tools. This research proposes use of Google's GTFS for a bus stop aggregation model (SAM) based on distance between individual stops, textual similarity, and common service areas. By measuring land use types within a specified service area based on SAM, this research helps in advancing our understanding of transit demand in the vicinity of bus stops. In addition, a systematic matching technique for aggregating stops (SAM) allows us to analyze the symmetry of boarding and alightings, which can observe a considerable passenger flow between specific time periods and symmetry by time period pairs (e.g., between AM and PM peaks) on an individual day. This research explores the potential generation of a time-varying O-D matrix from APC data, in conjunction with integrated land use and transportation models. This research aims at incorporating all valuable information - the time-varying alighting probability matrix (TAPM) that represents on-board passengers' trip purpose - into the O-D estimation process. A practical application is based on APC data on a specific transit route in the Minneapolis - St. Paul metropolitan area. This research can also provide other practical implications. It can help transit agencies and policy makers to develop decision-making tools to support transit planning, using improved databases with transit-related ADCS and parcel-level land use data. As a result, this work not only has direct implications for the design and operation of future urban public transport systems (e.g., more precise bus scheduling, improve service to public transport users), but also for urban planning (e.g., for transit oriented urban development) and travel forecasting.

The focus is on cross-sectional dependence in panel trade flow models. We propose alternative specifications for modeling time invariant factors such as socio-cultural indicator variables, e.g., common language and currency. These are typically treated as a source of heterogeneity eliminated using fixed effects transformations, but we find evidence of cross-sectional dependence after eliminating country-specific and time-specific effects. These findings suggest use of alternative simultaneous dependence model specifications that accommodate cross-sectional dependence, which we set forth along with Bayesian estimation methods. Ignoring cross-sectional dependence implies biased estimates from panel trade flow models that rely on fixed effects. / Series: Working Papers in Regional Science

Today’s metropolitan areas see changes in populations and land development occurring at faster rates than transportation planning can be updated. This dissertation explores the use of a new dataset from the location-based social networking spectrum to analyze origin-destination travel demand within Austin, TX. A detailed exploration of the proposed data source is conducted to determine its overall capabilities with respect to the Austin area demographics. A new methodology is proposed for the creation of origin-destination matrices using a peer-to-peer modeling structure. This methodology is compared against a previously examined and more traditional approach, the doubly-constrained gravity model, to understand the capabilities of both models with various friction functions. Each method is examined within the constructs of the study area’s existing origin-destination matrix by examining the coincidence ratios, mean errors, mean absolute errors, frequency ratios, swap ratios, trip length distributions, zonal trip generation and attraction heat maps, and zonal origin-destination flow patterns. Through multiple measures, this dissertation provides initial interpretations of the robust Foursquare data collected for the Austin area. Based upon the data analytics performed, the Foursquare data source is shown to be capable of providing immensely detailed spatial-temporal data that can be utilized as a supplementary data source to traditional transportation planning data collection methods or in conjunction with other data sources, such as social networking platforms. The examination of the proposed peer-to-peer methodology presented within this dissertation provides a first look at the potential of many-to-many modeling for transportation planning. The peer-to-peer model was found to be superior to the doubly-constrained gravity model with respect to intrazonal trips. Furthermore, the peer-to-peer model was found to better estimate productions, attractions, and zone to zone movements when a linear function was used for long trips, and was computationally more proficient for all models examined.

The focus is on cross-sectional dependence in panel trade flow models. We propose alternative specifications for modeling time invariant factors such as socio-cultural indicator variables, e.g., common language and currency. These are typically treated as a source of heterogeneity eliminated using fixed effects transformations, but we find evidence of cross-sectional dependence after eliminating country-specific effects. These findings suggest use of alternative simultaneous dependence model specifications that accommodate cross-sectional dependence, which we set forth along with Bayesian estimation methods. Ignoring cross-sectional dependence implies biased estimates from panel trade flow models that rely on fixed effects. / Series: Working Papers in Regional Science

The focus is on cross-sectional dependence in panel trade flow models. We propose alternative specifications for modeling time invariant factors such as socio-cultural indicator variables, e.g., common language and currency. These are typically treated as a source of heterogeneity eliminated using fixed effects transformations, but we find evidence of cross-sectional dependence after eliminating country-specific effects. These findings suggest use of alternative simultaneous dependence model specifications that accommodate cross-sectional dependence, which we set forth along with Bayesian estimation methods. Ignoring cross-sectional dependence implies biased estimates from panel trade flow models that rely on fixed effects. / Series: Working Papers in Regional Science

This research is concerned with the development of algorithmic approaches for solving problems that arise in the design and analysis of telecommunication networks, location-allocation distribution contexts, and intelligent transportation networks. Specifically, the corresponding problems addressed in these areas are a local access and transport area (LATA) network design problem, the discrete equal-capacity p-median problem (PMED), and the estimation of dynamic origin-destination path ows or trip tables in a general network. For the LATA network problem, we develop a model and apply the Reformulation-Linearization Technique (RLT) to construct various enhanced tightened versions of the proposed model. We also design efficient Lagrangian dual schemes for solving the linear programming relaxation of the various enhanced models, and construct an effective heuristic procedure for deriving good quality solutions in this process. Extensive computational results are provided to demonstrate the progressive tightness resulting from the enhanced formulations and their effect on providing good quality feasible solutions. The results indicate that the proposed procedures typically yield solutions having an optimality gap of less than 2% with respect to the derived lower bound, within a reasonable effort that involves the solution of a single linear program. For the discrete equal-capacity p-median problem, we develop various valid inequalities, a separation routine for generating cutting planes via specific members of such inequalities, as well as an enhanced reformulation that constructs a partial convex hull representation that subsumes an entire class of valid inequalities via its linear programming relaxation. We also propose suitable heuristic schemes for solving this problem, based on sequentially rounding the continuous relaxation solutions obtained for the various equivalent formulations of the problem. Extensive computational results are provided to demonstrate the effectiveness of the proposed valid inequalities, enhanced formulations, and heuristic schemes. The results indicate that the proposed schemes for tightening the underlying relaxations play a significant role in enhancing the performance of both exact and heuristic solution methods for solving this class of problems. For the estimation of dynamic path ows in a general network, we propose a parametric optimization approach to estimate time-dependent path ows, or origin-destination trip tables, using available data on link traffic volumes for a general road network. Our model assumes knowledge of certain time-dependent link ow contribution factors that are a dynamic generalization of the path-link incidence matrix for the static case. We propose a column generation approach that uses a sequence of dynamic shortest path subproblems in order to solve this problem. Computational results are presented on several variants of two sample test networks from the literature. These results indicate the viability of the proposed approach for use in an on-line mode in practice. Finally, we present a summary of our developments and results, and offer several related recommendations for future research. / Ph. D.