Aprendizagem por reforço multiagente (do inglês, MARL) é uma tarefa desafiadora em que agentes buscam, concorrentemente, uma política capaz de maximizar sua utilidade. Aprender neste tipo de cenário é difícil porque os agentes devem se adaptar uns aos outros, tornando o objetivo um alvo em movimento. Consequentemente, não existem garantias de convergência para problemas de MARL em geral. Esta tese explora um problema em particular, denominado escolha de rotas (onde motoristas egoístas deve escolher rotas que minimizem seus custos de viagem), em busca de garantias de convergência. Em particular, esta tese busca garantir a convergência de algoritmos de MARL para o equilíbrio dos usuários (onde nenhum motorista consegue melhorar seu desempenho mudando de rota) e para o ótimo do sistema (onde o tempo médio de viagem é mínimo). O principal objetivo desta tese é mostrar que, no contexto de escolha de rotas, é possível garantir a convergência de algoritmos de MARL sob certas condições. Primeiramente, introduzimos uma algoritmo de aprendizagem por reforço baseado em minimização de arrependimento, o qual provamos ser capaz de convergir para o equilíbrio dos usuários Nosso algoritmo estima o arrependimento associado com as ações dos agentes e usa tal informação como sinal de reforço dos agentes. Além do mais, estabelecemos um limite superior no arrependimento dos agentes. Em seguida, estendemos o referido algoritmo para lidar com informações não-locais, fornecidas por um serviço de navegação. Ao usar tais informações, os agentes são capazes de estimar melhor o arrependimento de suas ações, o que melhora seu desempenho. Finalmente, de modo a mitigar os efeitos do egoísmo dos agentes, propomos ainda um método genérico de pedágios baseados em custos marginais, onde os agentes são cobrados proporcionalmente ao custo imposto por eles aos demais. Neste sentido, apresentamos ainda um algoritmo de aprendizagem por reforço baseado em pedágios que, provamos, converge para o ótimo do sistema e é mais justo que outros existentes na literatura. / Multiagent reinforcement learning (MARL) is a challenging task, where self-interested agents concurrently learn a policy that maximise their utilities. Learning here is difficult because agents must adapt to each other, which makes their objective a moving target. As a side effect, no convergence guarantees exist for the general MARL setting. This thesis exploits a particular MARL problem, namely route choice (where selfish drivers aim at choosing routes that minimise their travel costs), to deliver convergence guarantees. We are particularly interested in guaranteeing convergence to two fundamental solution concepts: the user equilibrium (UE, when no agent benefits from unilaterally changing its route) and the system optimum (SO, when average travel time is minimum). The main goal of this thesis is to show that, in the context of route choice, MARL can be guaranteed to converge to the UE as well as to the SO upon certain conditions. Firstly, we introduce a regret-minimising Q-learning algorithm, which we prove that converges to the UE. Our algorithm works by estimating the regret associated with agents’ actions and using such information as reinforcement signal for updating the corresponding Q-values. We also establish a bound on the agents’ regret. We then extend this algorithm to deal with non-local information provided by a navigation service. Using such information, agents can improve their regrets estimates, thus performing empirically better. Finally, in order to mitigate the effects of selfishness, we also present a generalised marginal-cost tolling scheme in which drivers are charged proportional to the cost imposed on others. We then devise a toll-based Q-learning algorithm, which we prove that converges to the SO and that is fairer than existing tolling schemes.

Eine an den Bedürfnissen von Radfahrer*innen orientierte Planung ist für die Förde-rung des Radverkehrs unumgänglich. Um den Ausbau der Radverkehrsinfrastruktur ent-sprechend zu planen, sind jedoch Informationen zum Routenwahlverhalten der Radfah-renden und dessen Einflussfaktoren notwendig. Mit diesen Informationen können z.B. Wirkungen von Maßnahmen oder die zukünftige Nutzung von geplanter Radverkehrsinf-rastruktur abgeschätzt werden. Aus diesem Grund wurden vor allem in der letzten Dekade zahlreiche internationale Arbeiten vorangetrieben, die die Routenwahl auf Basis von GPS-basierten Daten zum real beobachteten Routenwahlverhalten untersuchten. Da die Studien meist für einzelne und nichtdeutsche Städte durchgeführt wurden, lassen sich die Ergebnisse jedoch nicht direkt auf den deutschen Raum übertragen. Um die Einflussfaktoren der Routenwahl für den deutschen Raum zu untersuchen, wird in dieser Arbeit die Routenwahl von Radfahrenden im Raum Dresden analysiert. Ba-sis der Untersuchung stellt ein während der Aktion STADTRADELN erhobener GPS-Daten-satz dar, der 18.459 Radfahrten von 1.361 Radfahrer*innen enthält und Auskunft über deren real beobachtetes Routenwahlverhalten gibt. Die erhobenen Routendaten wurden mit unterschiedlichen Verfahren aufbereitet und über ein Verkehrsnetz mit Sekundärda-ten angereichert, sodass die Eigenschaften der gewählten Routen berechnet und den je-weils erzeugten und nicht gewählten Routenalternativen gegenübergestellt werden konn-ten. Der Einfluss der unterschiedlichen Faktoren wurde anschließend mittels logistischer Regressionsanalyse untersucht. Das Ergebnis der Untersuchung zeigt auf, dass sich die folgenden Einflussfaktoren positiv auf die Routenwahl auswirken: + Existenz von Radverkehrsinfrastruktur + Geringe Längsneigungen + Gute Oberflächenbeschaffenheit (z.B. Asphalt) + Vorhandensein anderer Radfahrender entlang einer Route + Geringe zulässige Maximalgeschwindigkeit des motorisierten Verkehrs + Geringe Fahrstreifenanzahl des Kfz-Verkehrs + Durch Lichtsignalanlagen geregelte Knotenpunkte + Grün- und Wohngebietsflächen Demgegenüber üben die folgenden Einflussfaktoren einen negativen Einfluss auf die Wahl einer Route aus: – Zunehmende Distanz – Hohe Streckenanteile mit großer Längsneigung – Eine hohe maximale Längsneigung der Route – Geringe Breite der Radverkehrsführung – Höhere Unfallhäufigkeit entlang der Route Für einige Einflussfaktoren konnte kein (signifikanter) Einfluss auf die Routenwahl nachgewiesen werden. Dazu gehören bspw. die durchschnittlich tägliche Verkehrsstärke des Kfz-Verkehrs, der ruhende Verkehr oder die Häufigkeit von „rechts-vor-links“ geregel-ten Knotenpunkten. Die vorliegende Arbeit konnte dazu beitragen, die Wirkung der Einflussfaktoren auf die Routenwahl im Radverkehr einer deutschen Stadt zu quantifizieren. Eine Stärke der Untersuchung liegt in der Erhebung und Nutzung des umfangreichen GPS-Datensatzes zum real beobachteten Routenwahlverhalten der Radfahrer*innen. Das resultierende Routenwahlmodell kann für die Abschätzung von Maßnahmenwirkungen genutzt werden (z.B. im Rahmen der Verkehrsnachfragemodellierung). Eine kritische Reflexion der Ergebnisse und Methoden zeigt, dass die Qualität der Ana-lyse und ihrer Ergebnisse durch die Nutzung weiterer Sekundär- und Primärdaten sowie anderer Methoden weiter verbessert werden kann. Die Ergebnisse der Untersuchung geben einen Einblick in die Routenwahl der Radfah-renden des Untersuchungsgebiets. Im Rahmen der Arbeit konnte nicht geprüft werden, ob die Routenwahl der Radfahrer*innen in Dresden vergleichbar mit der Routenwahl in anderen deutschen Städten ist. Das entwickelte Verfahren kann jedoch auch für Analysen in anderen Städten angewendet werden, sodass verlässliche Aussagen zur Routenwahl deutscher Radfahrender in anderen deutschen Kommunen erarbeitet werden können. Es lassen sich insgesamt folgende Schlussfolgerungen ziehen: Um den Radfahrenden eine möglichst attraktive Radverkehrsinfrastruktur zu bieten und damit die Nutzung des Fahrrads zu fördern, sollte der Radverkehr auf eigener Radverkehrsinfrastruktur geführt werden, die möglichst breit ist und eine gute Oberflächenqualität besitzt. Entlang der Rad-routen sollten die Geschwindigkeiten des fließenden motorisierten Verkehrs möglichst reduziert werden. Die Bündelung von Radverkehrsströmen sowie die Führung durch Grünflächen kann zudem Fahrtkomfort und Sicherheitsempfinden erhöhen. / User-oriented bicycle planning is essential for the promotion of cycling in cities. In order to adequately plan and build cycling infrastructure, information on route choice be-haviour of cyclists and its influencing factors is essential. The knowledge about influencing factors of bicycle route choice can help to assess the effect of measures as well as future or potential utilisation of planned cycling infrastructure. Several international studies – particularly carried out in the last decade – examined route choice based on GPS-based revealed preference data. As the studies were mostly conducted for few and non-German cities the results, however, cannot be directly trans-ferred to the German area. In order to investigate bicycle route choice for the German area, the route choice of cyclists in the city of Dresden is analysed in this study. The basis of the investigation is a GPS data set collected during the CITY CYCLING campaign. It contains 18,459 bicycle trips by 1,361 cyclists and provides detailed information about the observed route choice be-haviour of cyclists. The collected route data were processed by using different methods. Linking the routes to a traffic supply network allowed enriching them with secondary data so that characteristics of the selected routes and the non-selected generated alternatives were determined. The influence of the different factors was investigated using logistic re-gression analysis. The results of the study reveal that the following influencing factors positively affect bicycle route choice: + Existence of cycling infrastructure + Low slopes + Good surface conditions (e.g. asphalt) + Presence of other cyclists along a route + Low maximum speed of motorised traffic + Few number of lanes for motorised traffic + Intersections controlled by traffic signals + Green and residential areas In contrast, the following factors show a negative influence on bicycle route choice: – Increasing distance – High proportion of routes with steep longitudinal gradients – A high maximum gradient along the route – Narrow width of the cycle lane – Higher accident frequency along the route For some influencing factors, no (significant) influence on route choice was found (e.g. for the average daily traffic volume of motorized traffic, stationary traffic or the frequency of right-of-way regulated intersections). The present study contributes to the quantification of influencing factors or rather their effect on bicycle route choice in a German city. Furthermore, the resulting route choice model can be used to estimate the effects of different measures (e.g. in the context of traffic demand modelling). A strength of the study is the extensive data set on revealed preferences of cyclists that has been used to analyse route choice behaviour. A critical reflection of the results and methods shows that the quality of the analysis and its results could further be improved by using more secondary and primary data as well as other models for analysis. The results of the study provide an insight into route choices of cyclists in the study area. However, it could not be determined whether route choice of cyclists in Dresden is similar to route choice of cyclists in other German cities. Nevertheless, the developed method could also be used to analyse route choice in other cities, so that reliable state-ments on route choice of German cyclists can be compiled. Overall, the following conclusions can be drawn: in order to provide cyclists the most attractive cycling infrastructure possible and, thus, promote the use of cycling in cities, bicycle traffic should be guided on dedicated and wide cycling infrastructure with good surface quality. Furthermore, the speed of moving motorised traffic should be reduced along the route segments if possible. Bundling of cycling traffic flows and routing cyclists through green spaces can furthermore increase riding comfort and the perception of safety.

In the last several years, there has been growing worldwide interest in making streets safer for all users--pedestrians, cyclists, and motorists. One approach, shared space, is a traffic calming technique as well as urban design concept. This technique strives to fully integrate the roadway into the urban fabric by removing elements such as lane markings, curbs, and traffic signs. By removing these elements and creating a more plaza-like space, these sites become ambiguous and no user group as priority. The technique is relatively new, and the majority of existing research concerns pedestrians only. This mixed methods research focused on six intersections in England with the goal of understanding how bicycle riders perceive and travel through shared space intersections. Using video observations of the six sites in three cities, three shared and three control, this project analyzed the variations in the paths cyclists rode through the intersections. Data were collected on several variables related to both the cyclists and their interactions with the site itself such as helmet use and riding through crosswalks. Path analysis required the development a new evaluative variable in order to compare individual paths by how much deviation there was in each path ridden as compared to other cyclists. Site-specific surveys addressed the perceptions, bicycling experience, demographics, and path and route preferences by cyclists at both shared space and control intersections. The analysis indicated that cyclists rode similarly through both shared and control intersections, and that a large percentage of riders preferred to ride farther from motor vehicles when given the space to do so. This project offered further insight in how to best design shared space projects for nonmotorized users by looking at the spatial layout and the elements that most influenced a rider’s path choice. Results indicated that, in these cases, shared space was not the panacea for nonmotorized users as some literature suggests, but nonetheless appeared to be a valid form of traffic calming. This research offered further insight in how to best design shared space projects for nonmotorized users by looking at the spatial layout and the elements that most influenced a rider’s path choice.

Value pricing is now an accepted strategy for congestion and demand management in metropolitan areas. Along with alternate congestion management strategies, many transportation agencies have started looking at value pricing as a method to help financial shortfalls of new congestion management projects. Value pricing allows revenue collected from toll facilities to reduce operational concerns with underutilized High Occupancy Vehicle (HOV) facilities and relieves environmental concerns by reducing travel demand. Recently, transportation agencies have become increasingly interested in a high-occupancy toll (HOT) lane value pricing system with time-dependent tolls or dynamic tolls that change by the congestion level. However, there is a lack of proper travel demand forecasting tools that can evaluate and determine the impacts of pricing on travelers' decision in relation to congestion. The current methods use aggregated and zonal based approaches that lack the capability of tracing individual travelers through the supply network in order to capture his/her travel decisions as it pertains to the estimated cost for toll usage. The conventional models do not consider individual traveler socio-economic characteristics, particularly the heterogeneous value of time (VOT). TRANSIMS (Transportation Analysis Simulation System) differs from current travel demand forecasting methods in its underlying concepts and structure. These differences include a consistent and continuous representation of time, a detailed representation of persons and households, time-dependent routing, and a person-based Microsimulator. The TRANSIMS Microsimulator is the only simulation tool that maintains the identity of the traveler throughout the simulation and is capable of accessing the database of each individual (e.g., income, age, trip purpose). It traces the movement of people as well as vehicles on a second-by-second basis. Although TRANSIMS environment has significantly improved over the past few years, there are still issues that need to be improved upon including: the pricing of a HOT lane with dynamic tolls and the rescheduling of activities (i.e., departure time choice model) in response to network conditions. The primary objectives of this study are to improve functions of TRANSIMS by modifying source codes in order to utilize non-linear, individual VOT function in route choice of a HOT lane value pricing system, to implement 15-min dynamic tolls that vary by level of service (i.e., volume/capacity ratio) in the HOT lane(s) and to develop departure time choice model. Testing the proposed methodologies using real-world data as case studies and evaluating the impacts of dynamic tolls and/or departure time choice model are other objectives of this study. The test site of the HOT lane system is a segment of I-5 northbound from Hwy 217 to I-405 near the central business district (CBD) in Portland metropolitan region, Oregon. The experimental analyses of the application of dynamic tolls and individual VOT demonstrate the feasibility of the proposed simulation methodology. The outputs from the microscopic analysis clearly indicate the effectiveness of the analysis in scrutinizing travelers' route choice behavior based on different socio-economic and travel characteristics when different toll rates are applied. The effects of individual VOT on route choice are consistent with intuition; that is, travelers with higher VOTs are more likely to choose the HOT lane(s). In addition, the impacts of various tolls on route choice are analyzed on the basis of socio-economic and trip characteristics of each traveler. In addition to the development of the dynamic value pricing along with individual VOT, the departure time choice model is also developed. The proposed method is a post-processing of route choice and represents a sequential decision making process of travelers who want to depart early or late based on congestion, individual attributes and activity characteristics. This paper presents the results of a departure time choice model and its impacts on a HOT lane system using Portland, Oregon as a case study. The results show that 13.9% of households did change their departure time because of congestion and/or tolls. / Ph. D.

No description available.

In dieser Arbeit wird ein simultanes Quell-, Ziel-, Verkehrsmittel- und Routenwahlmodell (Modell EVA-U) entwickelt, welches ein stochastisches Nutzergleichgewicht erreicht. Die Routenwahlmodelle der Verkehrsarten sind nicht mehr Teil der Umlegungsalgorithmen, sondern in das Nachfragemodell integriert. Dadurch ist eine konsistente Bewertung aller Alternativen (der Verkehrsarten) möglich. Das Simultanmodell EVA-U stellt eine Weiterentwicklung des Simultanmodells EVA von LOHSE dar. Das EVA-U-Modell ist den universalen Logit-Modellen zuzuordnen. Die Randsummenbedingungen der Verkehrsverteilung werden beachtet. Die Bewertung der Alternativen erfolgt mittels Generalisierter Kosten. Die Abhängigkeit von Routen wird berücksichtigt, ebenso die Tagesganglinie der Verkehrsnachfrage und die Fahrpläne des ÖV-Systems. Das Modell EVA-U erlaubt auch die Berücksichtigung von Routen intermodaler Verkehrsarten (z.B. P+R). Darüber hinaus ist die Integration eines Modells des ruhenden Verkehrs möglich. / In this thesis a simultaneous Trip Generation-, Distribution-, Modal-Split and Route Choice Model (modell EVA-U) is elaborated. The model tends to reach a stochastic user equilibrium. The route choice algorithms are not longer part of an assignment procedure but part of the demand model. A consistent assessment of properties of all transport systems is possible. The simultaneous model EVA-U is an advancement of the EVA-Model by Lohse. The model EVA-U is to be assigned to the generalised logit-models. All matrix constrains are taken into account. The assessment is effected by generalised costs. The dependence of routes is taken into account. Moreover, the integration of day time and the schedules of private transport lines is possible. Furthermore, it is possible to integrate a model of parked cars and circuits of inter-modal traffic forms (park and ride) in the Model EVA-U.

No description available.

Die Mobilität von Personen und Gütern, insbesondere in Städten, ist der Motor einer Volkswirtschaft. Dieser Motor kommt jedoch ins Stottern, wenn Staubildung im Stadtstraßennetzwerk einsetzt. Eine unvermeidbare Ursache von Staubildung stellen Verkehrsstörfälle dar, die schlimmstenfalls zu Gridlocks führen können. In der Folge werden hohe Kosten für Verkehr, Wirtschaft und Umwelt verursacht. Mit welchen Gegenmaßnahmen kann die Staubildung im Netzwerk effektiv bewältigt werden? Wie können entsprechende Gegenmaßnahmen realistisch noch vor einem praktischen Einsatz bewertet werden? Ausgehend von diesen Fragestellungen, widmet sich diese Dissertation der Entwicklung eines ereignisorientierten Routenwahlmodells für den Stadtstraßenverkehr und eines selbstorganisierten Störfallmanagements als Gegenmaßnahme zur Reduzierung negativer Auswirkungen der Staubildung. Zur Modellierung des Routenwahlverhaltens in ereignisreichen Stadtstraßennetzen wird das ereignisorientierte Routenwahlmodell entwickelt. Der Ausgangspunkt des Modells ist die diskrete Wahltheorie. Entscheidungsprozesse einzelner Autofahrer werden vor und während der Fahrt direkt simuliert. Der Entscheidungsprozess ist dabei maßgeblich von Beobachtungen lokaler Verkehrsbedingungen geprägt. Somit wird nachgebildet, dass Autofahrer flexibel auf unvorhergesehene Ereignisse durch Routenwechsel reagieren können. Auf diese Weise ist eine realistische Simulation des Routenwahlverhaltens von Autofahrern in der Stadt möglich. Das ereignisorientierte Routenwahlmodell ist zudem generisch formuliert. Es lässt sich zur Bewertung von Gegenmaßnahmen für störfallbedingte Staubildung einsetzen und bedient darüber hinaus ein breites Anwendungsspektrum. Der zweite Beitrag dieser Dissertation ist ein selbstorganisiertes Konzept für ein Störfallmanagement in Stadtstraßennetzen als Gegenmaßnahme zur Staubildung. Es vereint zwei lokal wirkende Prinzipien, deren Ausgangspunkte die Lichtsignalanlagen im Stadtnetzwerk sind. Mit verlängerten Rotzeiten werden Fahrzeuge an einer Kreuzung an der Einfahrt in einen Straßenabschnitt gehindert, wenn ein vorgesehener Rückstaubereich ausgeschöpft ist, da andernfalls Blockaden auf den Kreuzungen entstehen. Gleichzeitig werden noch freie Richtungen an der Kreuzung durch verlängerte Grünzeiten attraktiver gestaltet, um Autofahrer zum Umfahren der Staubildung zu motivieren. Die Anwendung der lokalen Wirkungsprinzipien stellt sich vollständig selbstorganisiert, d.h. ohne Vorgabe eines Planers, mit dem Ausmaß der Staubildung im Netzwerk ein. Simulationsstudien in zwei unterschiedlich komplexen Netzwerken haben die Machbarkeit des selbstorganisierten Störfallmanagements nachgewiesen. Gegenüber einem gewöhnlichen Netzwerk konnte für alle untersuchten Störfälle die Akkumulation zusätzlicher Fahrzeuge im Netzwerk während des Störfalls signifikant reduziert werden. / The mobility of people and goods, especially in urban areas, is of significant importance for national economies. However, recurrent congestion in urban road networks, caused by increased traffic demand, considerably restrains mobility on a daily basis. Another significant source of congestion are traffic incidents which even might lead to gridlock situations. Congestion raises high costs for traffic, economy and environment. Which countermeasures should be applied for an effective management of urban congestion? How can appropriate countermeasures be realistically evaluated? Based on these questions, this thesis is devoted to the development of an event-oriented route choice model for urban road traffic and a self-organized incident management strategy as an effective countermeasure for urban congestion. The first contribution of this thesis is an event-oriented route choice model for urban road networks. It is based on discrete choice theory and models decision-making processes of individual motorists before and during their journey. A key aspect of the proposed model is the motorist's ability to observe local traffic conditions. These observations are then included in the decision process. In this way, it can be modeled that motorists respond to unforeseen events by route revisions. This allows a realistic simulation of the route choice behavior of motorists in naturally eventful urban road networks. Furthermore, the event-oriented route choice model is flexibly formulated. It can be used for the evaluation of countermeasures for incident-related congestion and, moreover, allows a wide range of applications. The second contribution of this thesis is a self-organized concept of an incident management strategy in urban road networks as a countermeasure for urban congestion. It combines two locally acting principles on the basis of traffic lights in an urban road network. The inflow of vehicles into a road segment is regulated with restricted or skipped green times as soon as an allocated queuing capacity is depleted. Otherwise, blockages would result on the intersection. At the same time, yet free alternative directions are served with regular or even extended green times and, thus, might become more attractive to the driver than the original congested direction. The application of these local principles is realized in a completely self-organized manner, thereby scaling directly with the extent of congestion in the urban road network. Simulation studies in two networks with different complexity have proven the feasibility of the self-organized incident management. Compared to an ordinary network, the extents of additional vehicles due to investigated incidents were significantly reduced.

Dans les études sur le transport, les modèles de choix de route décrivent la sélection par un utilisateur d’un chemin, depuis son origine jusqu’à sa destination. Plus précisément, il s’agit de trouver dans un réseau composé d’arcs et de sommets la suite d’arcs reliant deux sommets, suivant des critères donnés. Nous considérons dans le présent travail l’application de la programmation dynamique pour représenter le processus de choix, en considérant le choix d’un chemin comme une séquence de choix d’arcs. De plus, nous mettons en œuvre les techniques d’approximation en programmation dynamique afin de représenter la connaissance imparfaite de l’état réseau, en particulier pour les arcs éloignés du point actuel. Plus précisément, à chaque fois qu’un utilisateur atteint une intersection, il considère l’utilité d’un certain nombre d’arcs futurs, puis une estimation est faite pour le restant du chemin jusqu’à la destination. Le modèle de choix de route est implanté dans le cadre d’un modèle de simulation de trafic par événements discrets. Le modèle ainsi construit est testé sur un modèle de réseau routier réel afin d’étudier sa performance. / In transportation modeling, a route choice is a model describing the selection of a route between a given origin and a given destination. More specifically, it consists of determining the sequence of arcs leading to the destination in a network composed of vertices and arcs, according to some selection criteria. We propose a novel route choice model, based on approximate dynamic programming. The technique is applied sequentially, as every time a user reaches an intersection, he/she is supposed to consider the utility of a certain number of future arcs, followed by an approximation for the rest of the path leading up to the destination. The route choice model is implemented as a component of a traffic simulation model, in a discrete event framework. We conduct a numerical experiment on a real traffic network model in order to analyze its performance.

La repr?sentation spatiale de l?aire de mod?lisation dans les mod?les de la demande de transports a peu chang? au cours des derni?res d?cennies. A cet ?gard, l??tat-de l?art repose encore largement sur le syst?me de centro?de-connecteur qui est utilis?e dans les mod?les classiques. Elle est une approche agr?g?e qui ignore la variabilit? physique li?e ? la dispersion des lieux d?sagr?g?s de r?sidence et d?activit? dans l?espace local. En cons?quence, le pouvoir explicatif des mod?les quant aux comportements de choix d?itin?raire et de mode demeure limit? ? l??chelle locale : Par exemple, la localisation d?sagr?g?e influence sur le choix entre une autoroute dont l??changeur est ?loign?, et un autre itin?raire non-autoroutier. Egalement, le rabattement terminal influence sur le partage modal auto vs. transports en commun. Nous pr?sentons une approche d?sagr?g?e pour la repr?sentation spatiale. Dans un d?coupage zonal, l?espace ? l?int?rieur d?une zone est repr?sent? de mani?re d?sagr?g?e stochastique. Pour chaque zone, les points d?ancrage sont d?finis relative aux n?uds du r?seau qui peuvent ?tre utilis?s pour acc?der au r?seau. Un itin?raire entre une paire de zones est ensuite consid?r? comme une chaine, compos?e de deux trajets terminaux, correspondants aux sections intrazonales de l?itin?raire, et d?un trajet principal correspondant ? la section entre deux points d?ancrage. En cons?quence, le mod?le de choix d?itin?raire est transform? ? un mod?le de choix conjoint d?une paire de point d?ancrage. Le vecteur des temps al?atoires terminaux est Normal Multidimensionnel donnant lieu ? un mod?le Probit de choix conjoint de points d?ancrage.Pour ?tendre au cadre multimodal, un mode collectif composite est d?fini comme une chaine compos?e des trois trajets modaux d?acc?s, principal, et de sortie, et les stations sont consid?r?es comme les points d?ancrage, connectant les trajets de rabattement au trajet principal. Un mod?le Logit Multinomial de choix de mode est estim? ? partir de l?Enqu?te Globale de Transport de 2001 pour le mode auto et le faisceau des modes collectifs composites, et est combin? avec les deux mod?les Probit correspondants au choix des stations / Spatial representation of modeling area in travel demand models has changed little over the course of last several decades. In this regard, the state-of-the-art still widely relies on the same centroid-connector system that has been used in classic models. In this approach continuum bidimensional space is lumped on centroids. It is an aggregate approach which ignores the physical variability linked to the scatteredness of disaggregate residence- and activity-places over the local space. Consequently the modeling performance in explaining route and mode choice behavior degrades at local scales: In route choice, disaggregate location influences the propensity between a distant interchange to a highway, or a nearby road. In mode choice, feeder service to public transportations influences the auto vs. transit modal share. We propose a disaggregate approach for spatial representation. Based on a zoning system, a stochastic disaggregate representation is used to characterize the space within a traffic analysis zone. For each zone, anchor-points are defined as the network nodes that are used for accessing to the network from within the local space. An itinerary between a pair of zones is then considered as a chain of legs composed of two terminal legs, corresponding to the intrazonal route sections, and one main leg between two anchor points. The route choice problem is transformed to a joint choice of a pair of anchor points. The vector of random terminal travel times is Multivariate Normal resulting in a Multinomial Probit model of choice of a pair of anchor points. To extend to the multimodal context, a transit composite mode is defined as a chain of access, main, and egress modal legs, and transit platforms are considered as anchor points connecting the feeder legs to the main line-haul leg. A Multinomial Logit mode choice model is estimated based on the 2001 Paris Household Travel Survey for the auto mode and the composite transit modes. It is joined with the two Multinomial Probit models corresponding to the choice of anchor points. The result is a joint model of mode and station choice with a disaggregate representation of the local space